Behavior Control
Cognitive Robotics
Rijeka 2018
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 2
Behavior Control needs hellip hellip integration of perception decisionplanning action on different complexity levels All parts depend on the others Improving one part may result in worse performance
Burkhard Cognitive Robotics Behavior Control 3
bull Household much more complicated than car driving bull Soccer much more complicated than chess
Programming Environments Different tools for bull Development of Programs bull Checking Programs bull Middleware Next Slides RobotControl developped for GermanTeam (Aibos)
Burkhard Cognitive Robotics Behavior Control 4
Widely used ROS (= Robot Operating System) httpwikirosorg
Burkhard 5 Cognitive Robotics Behavior Control
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 2
Behavior Control needs hellip hellip integration of perception decisionplanning action on different complexity levels All parts depend on the others Improving one part may result in worse performance
Burkhard Cognitive Robotics Behavior Control 3
bull Household much more complicated than car driving bull Soccer much more complicated than chess
Programming Environments Different tools for bull Development of Programs bull Checking Programs bull Middleware Next Slides RobotControl developped for GermanTeam (Aibos)
Burkhard Cognitive Robotics Behavior Control 4
Widely used ROS (= Robot Operating System) httpwikirosorg
Burkhard 5 Cognitive Robotics Behavior Control
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Behavior Control needs hellip hellip integration of perception decisionplanning action on different complexity levels All parts depend on the others Improving one part may result in worse performance
Burkhard Cognitive Robotics Behavior Control 3
bull Household much more complicated than car driving bull Soccer much more complicated than chess
Programming Environments Different tools for bull Development of Programs bull Checking Programs bull Middleware Next Slides RobotControl developped for GermanTeam (Aibos)
Burkhard Cognitive Robotics Behavior Control 4
Widely used ROS (= Robot Operating System) httpwikirosorg
Burkhard 5 Cognitive Robotics Behavior Control
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Programming Environments Different tools for bull Development of Programs bull Checking Programs bull Middleware Next Slides RobotControl developped for GermanTeam (Aibos)
Burkhard Cognitive Robotics Behavior Control 4
Widely used ROS (= Robot Operating System) httpwikirosorg
Burkhard 5 Cognitive Robotics Behavior Control
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Burkhard 5 Cognitive Robotics Behavior Control
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Burkhard 6 Cognitive Robotics Behavior Control
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Burkhard 7
Simulator Red robot in the middle Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
8
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Where to intercept the ball
Burkhard Cognitive Robotics Behavior Control 9
By calculation By simulation By learned behavior
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Which player can intercept first Based on calculation of intercept point
Burkhard Cognitive Robotics Behavior Control 10
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Pass to which team mate
Burkhard Cognitive Robotics Behavior Control 11
Based on calculations of intercepting players
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Simulation 2D RoboCup1997 Nagoya Final Match AT-Humboldt (Humboldt University of Berlin Germany) vs andhill (Tokyo Institute of Technology Japan)
Burkhard Cognitive Robotics Behavior Control 12
Burkhard Cognitive Robotics Behavior Control
Chess like Control for Soccer Evaluate options for future success
Choose the best alternative
13
Does not work for more Complicated situations
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 14
Classical Types of AgentRobot Behavior Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 15
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
Burkhard Cognitive Robotics Behavior Control
Reactive Behavior
environment sensors effectors
Immediate Action
agent
Perception
52 16
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 17
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 14
Classical Types of AgentRobot Behavior Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 15
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
Burkhard Cognitive Robotics Behavior Control
Reactive Behavior
environment sensors effectors
Immediate Action
agent
Perception
52 16
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 17
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Classical Types of AgentRobot Behavior Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 15
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
Burkhard Cognitive Robotics Behavior Control
Reactive Behavior
environment sensors effectors
Immediate Action
agent
Perception
52 16
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 17
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
Reactive Behavior
environment sensors effectors
Immediate Action
agent
Perception
52 16
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 17
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 17
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 18
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 19
Run for the ball
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Reactive (ldquostimulus-responserdquo)
Burkhard Cognitive Robotics Behavior Control 20
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
Goal directed behavior
environment sensors
Individual Plan
effectors
Individual Planner
Immediate Reaction
agent
Perception Action
57
Worldmodel
21
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 22
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 23
Acting according to a predefined goal
x
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Goal directed behavior
Burkhard Cognitive Robotics Behavior Control 24
Acting according to a predefined goal
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
ldquoMental Statesrdquo Past Belief (world model) Future Commitment (goal intention plan )
sensors
Commitments
effectors
Individual Planner
Immediate Reaction
agent
Belief
Perception Action
x
60 25
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
Cooperative Behavior
environment sensors effectors
Cooperative Planner
Individual Planner
Immediate Reaction
agent
Perception Action
63
Belief
Individual Commitments
Joint Commitments
26
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 27
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 28
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 29
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 30
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 31
Cooperation Joint intention (Double pass)
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Cooperative Behavior
Burkhard Cognitive Robotics Behavior Control 32
Cooperation Joint intention (Double pass)
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Control Architectures Distribution Interfaces Information flow
Burkhard Cognitive Robotics Behavior Control 33
Sensors Actuators
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
bdquoHorizontalldquo Structure
Burkhard Cognitive Robotics Behavior Control 34
Actuators
Sense Think Act
Sensors
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Synchronisation
Burkhard Cognitive Robotics Behavior Control 35
sense think act
sense think act
sense think act conflict
Sequential
Parallel
Real Time Requirements may lead to conflicts
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Different Complexities
Burkhard Cognitive Robotics Behavior Control 36
bull World Model
bull
bull Simple Percepts
bull Sensor Signals
bull Cooperative Planning
bull Planning
bull
bull Choice of Skill
bull Reaction (Stimulus Response)
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Layered Architecture Example
Burkhard Cognitive Robotics Behavior Control 37
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Deliberative Layer Long Term Planning Time consuming
Working Layer
Scheduling of bdquoskillsldquo Needs moderate time
Reactive Layer
Immediate reactions
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Layered Architecture Reaction Time
Burkhard Cognitive Robotics Behavior Control 38
Actuators
Sense Think Act
Sensors
Sense Think Act
Sense Think Act
Different cycle times at the layers
Time problems with ldquoupwards failuresrdquo 1 Problem on low level 2 Higher levels react with delay ldquomoment of schockrdquo
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Upwards Failure Planned behavior (double pass) Intercept fails but other player continues
Burkhard Cognitive Robotics Behavior Control 39
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
How to Organize Data Flow
Burkhard Cognitive Robotics Behavior Control 40
Sense Think Act
Sensors Actuators
Signals
WorldModel
Response
Planning
Action
Plan
Need for reduction of dependencies
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Classical One-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 41
Sense Think Act
Sensors Actuators
Worldmodel
Beliefs
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 42
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Low-level Controller
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Classical Two-Pass-Architecture
Burkhard Cognitive Robotics Behavior Control 43
Sense Think Act
Sensors Actuators
Worldmodel
Knowledge Base
Pilot
Navigator
Mission Planner
Low-level Contoller
Example 3-Tiered (3T) Architecture (NASA)
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 44
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard
Concept (Bounded) Rationality Rational Choice Assumption bull Agents act as utility maximizers Critics (Simon) Bounded Rationality bull Only limited knowledge about real world available bull Only limited resources for deliberation
Needs exact knowledge about future consequences
Cognitive Robotics Behavior Control 45
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Ideal Rational Agents
Aspects of the definition bull Performance measure determines the purposes of agentrobot bull Design problem designer has to built the necessary means
Critical assumption Real agents are not ldquoidealrdquo rational agents
Burkhard Cognitive Robotics Behavior Control 46
Definition by RussellNorvig Artificial Intelligence ndash A Modern Approach
For each possible percept sequence an ideal rational agent should do whatever action is expected to maximize its performance measure on the basis of the evidence provided by the percept sequence and whatever built-in knowledge the agent has
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Limitations of Rational Agents Limitations of ldquoidealrdquo agent bull ldquoevidence provided by the percept sequencerdquo limited by partial and incomplete information (belief) bull ldquowhatever built-in knowledgerdquo limited by insufficient internal models about the world Further limitations of ldquorealrdquo agent bull ldquomaximize its performance measurerdquo limited by available resources
bull hardwaresoftware bull time constraints
Burkhard 47 Cognitive Robotics Behavior Control
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Bounded Rationality
Burkhard Cognitive Robotics Behavior Control 48
Real agents are neither ldquoidealrdquo nor completely ldquorationalrdquo agents Compromises by design bull Dealing with trade-offs bull Do ldquoas best as canrdquo
ldquoBounded Rationalityrdquo Efficiency wrt limited resources
Notion from economics introduced by Herbert Alexander Simon 1955 (further aspect Irrationalities of decisions)
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 49
Player at time t
Ball observed at time t
Timepoint t
Intercept expected at time t+9
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 50
Player at time t
Timepoint t+3
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+9
Ball observed at time t
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 51
Player at time t
Timepoint t+6
Player at time t+3
Ball observed at time t+3
Intercept expected at time t+11
Ball observed at time t
Player at time t+6
Ball observed at time t+6
Time lost by changing directions
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
Conflicts between oldnew Options Keep old option
+ Stability + Reliabilty (cooperation) - Fanatism (misses better options)
Change for new option
+ Adapt to better options - May lead to oscillations
Treatment of conflict is up to choice by designer (architecture may even cause an implicit design decision)
52
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Protocols for Coordination Coordination possible without communication hellip hellip if all robots have the same world model and follow the same protocol Otherwise communication can help to bull unify world model bull distribute rolestasks (by protocol or negotiation or leader hellip)
Burkhard Cognitive Robotics Behavior Control 53
Communication - needs time - can be disrupted - can be inconsistentconflicting
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Example Roles by Protocol
Burkhard Cognitive Robotics Behavior Control 54
Role Task Assignment by Goalie defend goal fixed Attacker ball handling closest to ball Supporter support attacker close to attacker Defender backward support most back
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Example Stability vs Adaptation
Burkhard Cognitive Robotics Behavior Control 55
Solution Keep role in case of small deviations (ldquoHysteresis controlrdquo)
Role change Player closest to ball is attacker Distance to ball can oscillate by noisy observations
Difference of distances to ball
attacker
supporter
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Competing Desires Robot wants eg to - Change position (for supporting) - Avoid obstacles - Look for landmarks (for localization) - Observe the ball - Observe other players Can he pursue all these desires in parallel Rational behavior Commit only to achievable intentions
Burkhard Cognitive Robotics Behavior Control 56
Possible solution ldquoScreen of admissibilityrdquo Adapt new intentions only if not in conflict with already
adapted intentions (reg gives priority to stability)
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Least Commitment Cannot plan all future details in advance
Burkhard Cognitive Robotics Behavior Control 57
Solution ldquoLeast commitmentrdquo Postpone decisions as long as possible
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 58
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
BDI Agent Architecture Most popular architecture for reasoning agents Originally based on concepts of Michael E Bratman ldquoIntention Plans and Practical Reasonrdquo 1987 The architecture is built on - Possible facts about the world - Potential options the agent might achieve BDI stands for Beliefs Information the agent has about the world Desires States of affairs the agent would like to accomplish Intentions States of affairs the agent is trying to accomplish
Burkhard 59 Cognitive Robotics Behavior Control
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control
BDI-Modell Belief (world modell) Desire (useful options) Intention (committed options)
perceive Belief
select Desire
means-ends Intention
execute sense
new_Belief = update(Perception old_Belief) new_Desires = select (new_Beliefold_Desires) new_Intentions = means-ends(new_Belief new_Desires _old_Intentions)
61 60
Consistency ndashDesires may be inconsistent ndashIntentions must be consistent
Intentions set a screen of admissibility Only those desires may be adopted which are consistent with recent intentions
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
AgentSpeak and Jason AgentSpeak is a logic-based agent-oriented programming language implementing (some) concepts of BDI-architectures proposed by Arnand S Rao 1996 based on experiences with
PRS (Georgeff Lansky 1987) dMARS (Kinny 1993) Agent-0 (Shoham 1993)
Jason extension of AgentSpeak with Prolog-like syntactic structures interpreter in Java highly customizable developped by Rafael H Bordini Jomi F Huumlbner and others
Burkhard 61 Cognitive Robotics Behavior Control
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Interpreter
Works with Plan Library (initially filled) Belief Base (Memory of actual beliefs) Event Base (Memory for changes of beliefs and goals) Intention Base (Stacks of pending goals) Selection functions to select from the different Bases
Burkhard Cognitive Robotics Behavior Control 62
Next Slides Syntax and Informal Semantics Cited from BordiniHubner Jason - A Java-based interpreter for an extended version of AgentSpeak Release Version 095 February 2007 httpjasonsourceforgenetJasonpdf
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Beliefs in Jason
Beliefs first-order formulae ball(10 10) agent believes the ball is at position (10 10)
Beliefs can have annotations ball(10 10)[source(percept)] information was perceived from environment
Support for strong negation (besides negation by CWA) ~near(ball) agent believes itrsquos not near the ball
Belief base can also process (Prolog-like) rules
Burkhard 63 Cognitive Robotics Behavior Control
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Goals
2 types of goals Achievement goals for calling plans
kick(ball) might eg invoke a plan to bend the knee and kick
Test goals for tests of beliefs see(ball) succeeds if the agent actually sees the ball
Burkhard 64 Cognitive Robotics Behavior Control
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Plans
Plan in rule form triggeringEvent (Change of beliefs goals hellip) context (Conditions which must hold) lt- body (Sequence of goals and external actions)
Burkhard 65 Cognitive Robotics Behavior Control
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Simplified Reasoning Cycle
1 Update belief base by external percepts 2 Update event base according to previous steps 3 Select actual triggering event e 4 Determine relevant plans by unification with e 5 Determine applicable plans by checking contexts 6 Select a plan p from applicable plans 7 Update actually processed intention according to p resp initialize new intention (for external event) 8 Select intention i for processing 9 Execute next subgoal from top of intention i perform external action or update belief or test belief
Burkhard 66 Cognitive Robotics Behavior Control
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Implementation of Soccer Agents
67 Burkhard Cognitive Robotics Behavior Control
Perceptors Effectors Control (ldquoAgentrdquo)
RoboCup Simspark
Player-Program
Player
each 20 msec bullJoint angles bullAcceleration bullMicrophon bullcamera) hellip
Motor commands Loudspeaker hellip
) Camera percepts only each 60 msec ldquo
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Implementation of Soccer Agents Sense-think-act-cycle Sense process perceptor data from SimSpark Simulator implemented in Java Think analyse sitation and specify goals implemented in Jason Act send action commands to SimSpark Simulator implemented in Java
Burkhard Cognitive Robotics Behavior Control 68
Experimental Implementation of Soccer Agents by Dejan Mitrovic (Novi Sad)
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Example of a Simple Agent
Burkhard 69 Cognitive Robotics Behavior Control
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
(Partial) Implementation of the Example
Burkhard 70 Cognitive Robotics Behavior Control
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
DPA = Double Pass Architecture hellip another approach to implement BDI
Burkhard Cognitive Robotics Behavior Control 71
Diploma Thesis Ralf Berger used in RoboCup 2D league
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 72
1 Trial (bdquoChess-likeldquo)
bull Foresight simulation
bull Choice of best alternative
Result Useful only for short term decisions
2 Trial (bdquoEmergenceldquo)
If every player behaves in some simple optimal way
then a double pass can emerge without planning Result Double pass emerges only from time to time
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
How to program a double pass
Burkhard Cognitive Robotics Behavior Control 73
3 Trial
bull Use Bratmanacutes concept of bdquobounded rationalityldquo
Belief-Desire-Intention-Architecture (BDI)
bull Use Case-Based Reasoning
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Only partial plan in the beginning
Burkhard Cognitive Robotics Behavior Control 74
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute10acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 75
Analysis of situation ltsituation descriptiongt
Dribbling on path ltgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute11acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 76
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt to team mate 10
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute10acuteacute
Dribbling on path ltgt 5acute11acuteacute
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute12acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 77
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltgt
5acute10acuteacute
5acute12acuteacute
Run on path ltgt over opponent 7
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute13acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 78
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute14acuteacute(Least Commitment)
Burkhard Cognitive Robotics Behavior Control 79
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute14acuteacute Run on path ltgt
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute15acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 80
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
Run to ball on path ltgt kicked by team mate 10
5acute15acuteacute Run on path ltgt
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute16acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 81
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute16acuteacute Run on path ltgt
Run to ball on path ltgt kicked by team mate 10
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute17acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 82
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
Intercept ball at point ltgt optimal intercept point
5acute17acuteacute Run to ball on path ltgt
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute18acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 83
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute 5acute18acuteacute
Intercept ball at point ltgt optimal intercept point Run to ball on path ltgt
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute19acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 84
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltgt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Time 5acute20acuteacute (Least Commitment)
Burkhard Cognitive Robotics Behavior Control 85
Analysis of situation ltsituation descriptiongt
Dribbling on path ltpath parametersgt
Kick with parameters ltball speed vectorgt
Run on path ltpath parametersgt
Intercept ball at point ltpositiongt
Run to ball on path ltpath parametersgt
5acute10acuteacute
5acute12acuteacute 5acute13acuteacute
5acute17acuteacute
5acute19acuteacute 5acute20acuteacute
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Hierarchy of Options
Burkhard Cognitive Robotics Behavior Control 86
PlaySoccer
Offensive Defensive
Score OffsideTraAttack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Result of ldquoDeliberatorldquo Intention Subtree
Burkhard Cognitive Robotics Behavior Control 87
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 88
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 89
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
bdquoExecutorldquo checks conditions on all levels of the hierarchy and decides by least commitment principle
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 90
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 91
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 92
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 93
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Activity Path Present state of an Intention
Burkhard Cognitive Robotics Behavior Control 94
PlaySoccer
Offensive Defensive
Score OffsideTrap Attack ChangeWings1 DoublePass2 DoublePass1
Dribble
Pass
Intercept
Run
Kick
Reposition
Pass ready next Run
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Double-Pass Architecture bull Predefined Option Hierarchy bull Deliberator bull Executor
bdquoDoubledldquo 1-Pass-Architecture
1 Pass Deliberator (goal-oriented intention subtree) 2 Pass Executor (stimulus-response activity path)
- on all levels - Differences to ldquoclassicalrdquo Programming
Control flow by Deliberation (ldquoAgent- orientedrdquo) Runtime organization by 2 Passes through all levels
Burkhard Cognitive Robotics Behavior Control 95
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Overview Introduction Control Architectures Aspects of Rationality BDI Architectures Behavior Based Robotics
Burkhard Cognitive Robotics Behavior Control 96
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 97
Behavior Based Robotics Simple behavior by eg - Immediate reaction to sensor data (sensor-actor-coupling) - Simple physical bdquotransformationldquo (clever design) Intelligent action without intelligent thinking bull No worldmodel bull No symbols bull No deliberation
Hypothesis Complex behavior emerges by combination of simple behaviors
Emergent behavior Complex behavior emerges by interaction of situated robots with the environment
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 98
bdquoNew AIldquo Since middle of 1980s Orientation on natural principles - Emergent behavior - Situated agentsrobots - No internal representation
Papers by Rodney Brooks bdquoElefants donacutet play chessldquo bdquoIntelligence without reasonldquo bdquoIntelligence without representationldquo Patty Maes
Kismet
Coq Roomba
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Critics on Classical AI GOFAI = ldquoGood Old Fashioned AIrdquo Problems with bull Closed world assumption bdquoeverything is knownldquo bull Frame problembdquoall assumptionseffects are modelled bull Physical systems hypothesis complete symbolic
representation
Burkhard Cognitive Robotics Behavior Control 99
1966-72 Robot Shakey (Stanford) with hierarchical planner STRIPS (bdquoStanford Research Institute Problem Solverldquo)
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 100
Physical Symbol System Hypothesis
A physical symbol system has the necessary and sufficient means for intelligent actionldquo
NewellSimon Computer Science as Empirical Inquiry Symbols and Searchldquo
Many critics (Dreyfus Searle Penrose Brooks Maes Pfeiffer)
Needs bull Complete Descriptions of the Worlds bull Algorithms for actions
GOFAI= bdquogood old fashioned AIldquo
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 101
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 102
Physical Grounding Hypothesis This hypothesis states that to build a system that is intelligent it is necessary to have its representations grounded in the physical world Our experience with this approach is that once this commitment is made the need for traditional symbolic representations fades entirely The key observation is that the world is its own best model It is always exactly up to date It always contains every detail there is to be known The trick is to sense it appropriately and often enough
To build a system based on the physical grounding hypothesis it is necessary to connect it to the world via a set of sensors and actuators Typed input and output are no longer of interest They are not physically grounded RA Brooks Elephants Donacutet Play Chess
New Problem
But To bring the Beer from the basement the robot should have an idea about the location etc
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 103
Subsumption Architecture (Brooks) Example
Sense Act
Sensors Actuators
Avoid obstacles
Drive forward
Collect can
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Burkhard Cognitive Robotics Behavior Control 104
Subsumption Architecture (Brooks) bull Behaviors realized by simple AFSM (augmented finite state machines) bull No other internal modelling bull Layers Hierarchical collection of behaviors bull Parallel control by all layers bull In case of conflicts higher layer overwrights (bdquosubsumesldquo) other layers
First successful robot designs for simple tasks
Problems with too many behaviors Design and prediction of resulting behavior
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-
Consequence Different Approaches Needed Reactive Behavior
like Stimulus-Response short term bdquosimpleldquo behavior patterns simple skills
Deliberative Behavior Goal directed plan based behavior long term bdquocomplexldquo behavior
Hybrid Combination of reactive and deliberative behavior eg goal driven usage of reactive skills
Burkhard Cognitive Robotics Behavior Control 105
In robotics up to now More emphasis put to aspects of low level control Recently Increasing interest in high level control
- Behavior Control
- Overview
- Behavior Control needs hellip
- Programming Environments
- Foliennummer 5
- Foliennummer 6
- Foliennummer 7
- Chess like Control for Soccer
- Where to intercept the ball
- Which player can intercept first
- Pass to which team mate
- Simulation 2D
- Chess like Control for Soccer
- Overview
- Classical Types of AgentRobot Behavior
- Reactive Behavior
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Reactive (ldquostimulus-responserdquo)
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- Goal directed behavior
- ldquoMental Statesrdquo
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Cooperative Behavior
- Control Architectures
- bdquoHorizontalldquo Structure
- Synchronisation
- Different Complexities
- Layered Architecture Example
- Layered Architecture Reaction Time
- Upwards Failure
- How to Organize Data Flow
- Classical One-Pass-Architecture
- Classical Two-Pass-Architecture
- Classical Two-Pass-Architecture
- Overview
- Concept (Bounded) Rationality
- Ideal Rational Agents
- Limitations of Rational Agents
- Bounded Rationality
- Example Stability vs Adaptation
- Stability vs Adaptation
- Stability vs Adaptation
- Conflicts between oldnew Options
- Protocols for Coordination
- Example Roles by Protocol
- Example Stability vs Adaptation
- Competing Desires
- Least Commitment
- Overview
- BDI Agent Architecture
- BDI-Modell
- AgentSpeak and Jason
- Interpreter
- Beliefs in Jason
- Goals
- Plans
- Simplified Reasoning Cycle
- Implementation of Soccer Agents
- Implementation of Soccer Agents
- Example of a Simple Agent
- (Partial) Implementation of the Example
- DPA = Double Pass Architecture
- How to program a double pass
- How to program a double pass
- Only partial plan in the beginning
- Time 5acute10acuteacute (Least Commitment)
- Time 5acute11acuteacute (Least Commitment)
- Time 5acute12acuteacute (Least Commitment)
- Time 5acute13acuteacute (Least Commitment)
- Time 5acute14acuteacute(Least Commitment)
- Time 5acute15acuteacute (Least Commitment)
- Time 5acute16acuteacute (Least Commitment)
- Time 5acute17acuteacute (Least Commitment)
- Time 5acute18acuteacute (Least Commitment)
- Time 5acute19acuteacute (Least Commitment)
- Time 5acute20acuteacute (Least Commitment)
- Hierarchy of Options
- Result of ldquoDeliberatorldquo Intention Subtree
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Activity Path Present state of an Intention
- Double-Pass Architecture
- Overview
- Behavior Based Robotics
- bdquoNew AIldquo
- Critics on Classical AI
- Physical Symbol System Hypothesis
- Physical Grounding Hypothesis
- Physical Grounding Hypothesis
- Subsumption Architecture (Brooks)
- Subsumption Architecture (Brooks)
- Consequence Different Approaches Needed
-