minfaculty departmentofinformatics - introductiontorobotics€¦ · minfaculty...

Universität HamburgDER FORSCHUNG | DER LEHRE | DER BILDUNG

MIN FacultyDepartment of Informatics

- Introduction to Robotics

Introduction to RoboticsLecture 13

Jianwei Zhang, Lasse Einig[zhang, einig]@informatik.uni-hamburg.de

University of HamburgFaculty of Mathematics, Informatics and Natural SciencesDepartment of InformaticsTechnical Aspects of Multimodal Systems

July 08, 2016

J. Zhang, L. Einig 456



Architectures of Sensor-based Intelligent Systems Introduction to Robotics

Outline

IntroductionKinematic EquationsRobot DescriptionInverse Kinematics for ManipulatorsDifferential motion with homogeneous transformationsJacobianTrajectory planningTrajectory generationDynamicsRobot Control





Outline (cont.)

Task-Level Programming and Trajectory GenerationTask-level Programming and Path PlanningTask-level Programming and Path PlanningArchitectures of Sensor-based Intelligent Systems

The CMAC-ModelThe Subsumption-ArchitectureControl Architecture of a FishProcedural Reasoning SystemBehavior FusionHierarchyArchitectures for Learning Robots





Architectures of Sensor-based Intelligent Systems

OverviewI Basic behaviorI Behavior fusionI SubsumptionI Hierarchical architecturesI Interactive architectures





The Perception-Action-Model with Memory

perception behaviors

memory

actionsensors




Architectures of Sensor-based Intelligent Systems - The CMAC-Model Introduction to Robotics

CMAC-Model

CMAC: Cerebellar Model Articulation Controller

S sensory input vectors (firing cell patterns)A association vector (cell pattern combination)P response output vector (A ·W )W weight matrix

The CMAC model can be viewed as two mappings:

f : S −→ A

g : A W−→ P





CMAC-Model (cont.)

S1

S2

S3

S4

S5

A1

A2

A3

A4

A5

A6

R1

R2

R3

W1,1

W1,2

W1,3

W1,4

W1,5

W1,6

W2,1

W2,2

W2,3

W2,4

W2,5

W2,6

W3,1

W3,2

W3,3

W3,4

W3,5

W3,6

R1

R2

R3

sensory cells association cells adjustable weight response cells





B-Spline-Model

The B-Spline model is an ideal implementation of theCMAC-Model. The CMAC model provides an neurophysiologicalinterpretation of the B-Spline model.





Alvinn – Visual Navigation

CMU – Carnegie Mellon UniversityJ. Zhang, L. Einig 464




Action-oriented Perception

Percept-1 Behavior-1 Response-1Percept-2 Behavior-2 Response-2Percept-3 Behavior-3 Response-3

Fusion

Percept-1

Percept-2

Percept-3

Percept Behavior Response

Percept-1

Percept-2

Percept-3

Behavior Responseone of




Architectures of Sensor-based Intelligent Systems - The Subsumption-Architecture Introduction to Robotics

The Subsumption ArchitectureI hierarchical structure of behaviorI higher level behaviors subsumpt lower level behaviors

Modeling/planning paradigm

I program composed of sequence of stepsI functional units transform sensory data snapshots to actionsI series of actions achieve a specified goal

Subsumption architecture

I e.g. the mobile robot Rug WarriorI begins with cruise behaviorI moves forward

I a follow behavior subsumpts the cruise behaviorI triggered by photo cellsI moves towards the light

I avoid behavior suppresses follow and cruiseI infrared sensors detect imminent collisionI attempts to avoid the obstacles

I escape behavior uses bumper sensor to sense collisionsI reverses and bypasses the obstacleI top-level behavior, detect sound pattern, makes the robot play a

tuneI detect sound pattern subsumpts all lower level behaviors by

stopping or following the sound pattern

cruise

follow

avoid

escape

detect sound pattern

S

S

S S

photo cells

IR detector

bumper

microphone piezo buzzer

motors

[18]





Foraging and FlockingI multi-robot architectureI basic behaviors are sequentially executed

I composite behaviors built from basicsI by vector summationI hierachical composite behaviorsI include composite behaviors

homing

dispersion

aggregation

safe wandering

following

basic behaviors composite behaviors

sensory inputs/sensory conditions

actuator output

flocking

foraging=wandering+dispersion+following+homing+flocking

herding=wandering+surrounding+flockingsurrounding=wandering+following+aggregationflocking=wandering+aggregation+dispersion

[19]J. Zhang, L. Einig 467




Cockroach Neuron / Behaviors

ringfrequency

synapscurrents currents

R Ccell membrane

thresholdvoltage

SENSORS BEHAVIORS

ingestion

findingfood

edgefollowing

wandering

mouthtactile

mouthchemical

antennachemical

antenna




Architectures of Sensor-based Intelligent Systems - Control Architecture of a Fish Introduction to Robotics

Control Architecture of a Fish

Control and information flow in artificial fishPerception sensors, focuser, filterBehaviors behavior routinesBrain/mind habits, intention generatorLearning optimizationMotor motor controllers, actuators/muscles




Architectures of Sensor-based Intelligent Systems - Control Architecture of a Fish Introduction to Robotics

Control Architecture of a Fish (cont.)




Architectures of Sensor-based Intelligent Systems - Procedural Reasoning System Introduction to Robotics

Procedural Reasoning System

INTERPRETERMONITOR COMMANDGENERATOR

OPERATORINTERFACE

plansdesires intentionsbeliefs

actuatorssensors [20]




Architectures of Sensor-based Intelligent Systems - Behavior Fusion Introduction to Robotics

Hierarchical Fuzzy-Control of a Robot

subgoalapproaching

local collisionavoidance

commandfrom others

situationevaluation

knowledge base

worldmodel

fuzzy controller

subgoal planning

robot perception

replanningsubgoal sequence

commandingof other robots,human users

[21]




Architectures of Sensor-based Intelligent Systems - Behavior Fusion Introduction to Robotics

Behavior FusionFuzzy rules evaluate current situation.Situation evaluation determines 3 fuzzy-parameters

I the priority K of the LCA rule baseI the replanning selectorI NextSubgoal (whether a subgoal has been reached)

Typical rule IF (SL85 IS HIGH) AND (SL45 IS VL) AND (SLR0 IS VL) AND(SR45 IS VL) AND (SR85 IS VL) THEN (Speed IS LOW) AND (Steer ISPM) K IS HIGH AND Replan IS LOW

Translation If the leftmost proximity sensor detects an obstaclewhich is near and the other sensors detect no obstacle atall, then steer halfway to the right at low speed. Mainlyperform obstacle avoidance. No re-planning required.

Coordination of multiple rule basesSpeed = SpeedLCA · K + SpeedSA · (1− K )Steer = SteerLCA · K + SteerSA · (1− K )

LCA: Local Collision Avoidance SA: Subgoal Approach




Architectures of Sensor-based Intelligent Systems - Hierarchy Introduction to Robotics

Hierarchy

Real-Time Control System (RCS)I RCS reference model is an architecture for intelligent systems.I Processing modes are organized such that the BG (Behavior

Generation) modules form a command tree.I Information in the knowledge database is shared between WM

(World Model) modules in nodes within the same subtree.

[22]

Examples of functional characteristics of the BG and WM modules:





Hierarchy (cont.)





Hierarchy (cont.)

G1 M1 H1

G2 M2 H2

G3 M3 H3

G4 M4 H4

from sensors to servos

to higher levelsof interpretation

from a prioriknowledge base

from higher levelsof control

parameters ofsensed world

parameters ofexpected world

specification ofworld model error

parameters ofworld model

parameters of higherlevel world model

specification ofsub-task

[22]





Sensor-Hierarchy

Level 0

Level I

Level II

Level III

Level IV

raw data

Properties of points in space

Descriptions of aggregatesof points and their features(object elements)

Recognition of aggregates offeatures (objects)

Descriptions of relations ofobjects or unrecognized aggre-gates

Recognition of relations ofobjects

[22]





Other examples

[23]




Architectures of Sensor-based Intelligent Systems - Architectures for Learning Robots Introduction to Robotics

An Architecture for Learning Robots

[24]





AuRA Architecture

schema controllermotor perceptual

REPRESENTATION

learning

plan recognitionuser profile

spatial learning

opportunism

on-lineadpation

user input

user intentions

spatial goals

missionalterations

teleautonomy

mission planner

spatial reasoner

plan sequencer

Actuation Sensing

HiearchicalComponent

ReactiveComponent

[25]J. Zhang, L. Einig 481




Atlantis Architecture

CONTROL

SEQUENCING

DELIBERATIVE

SENSORS ACTUATORS

statusactivation

invocationresults

[26]





RACERobustness by Autonomous Competence Enhancement

Blackboard

HTN PlannerOWLOntology

ExecutionMonitor

Robot

Capabil-ities

Conceptualizer

Reasoningand

Interpretation

ExperienceExtractor/Annotator

UserInterface

Perception

PerceptualMemory

OWLconcepts

OWLconcepts

new concepts

experiences

fluents

fluentsfluents

expe-riences instructions

instructions

fluents

inital state,goal plan

plan,goal

fluents,schedule

planROSactions

actionresults

continuousdata

sensor data

[27]





[1] K. Fu, R. González, and C. Lee, Robotics: Control, Sensing, Vision,and Intelligence.McGraw-Hill series in CAD/CAM robotics and computer vision,McGraw-Hill, 1987.

[2] R. Paul, Robot Manipulators: Mathematics, Programming, andControl : the Computer Control of Robot Manipulators.Artificial Intelligence Series, MIT Press, 1981.

[3] J. Craig, Introduction to Robotics: Pearson New InternationalEdition: Mechanics and Control.Always learning, Pearson Education, Limited, 2013.

[4] J. F. Engelberger, Robotics in service.MIT Press, 1989.

[5] W. Böhm, G. Farin, and J. Kahmann, “A Survey of Curve andSurface Methods in CAGD,” Comput. Aided Geom. Des., vol. 1,pp. 1–60, July 1984.





[6] J. Zhang and A. Knoll, “Constructing fuzzy controllers with B-splinemodels-principles and applications,” International Journal ofIntelligent Systems, vol. 13, no. 2-3, pp. 257–285, 1998.

[7] M. Eck and H. Hoppe, “Automatic reconstruction of b-splinesurfaces of arbitrary topological type,” in Proceedings of the 23rdAnnual Conference on Computer Graphics and InteractiveTechniques, SIGGRAPH ’96, (New York, NY, USA), pp. 325–334,ACM, 1996.

[8] M. C. Ferch, Lernen von Montagestrategien in einer verteiltenMultiroboterumgebung.PhD thesis, Bielefeld University, 2001.

[9] N. J. Nilsson, “A mobile automaton: An application of artificialintelligence techniques,” tech. rep., DTIC Document, 1969.

[10] J. H. Reif, “Complexity of the mover’s problem and generalizationsextended abstract,” Proceedings of the 20th Annual IEEE





Conference on Foundations of Computer Science, pp. 421–427,1979.

[11] J. T. Schwartz and M. Sharir, “A survey of motion planning andrelated geometric algorithms,” Artificial Intelligence, vol. 37, no. 1,pp. 157–169, 1988.

[12] J. Canny, The complexity of robot motion planning.MIT press, 1988.

[13] T. Lozano-Pérez, J. L. Jones, P. A. O’Donnell, and E. Mazer,Handey: A Robot Task Planner.Cambridge, MA, USA: MIT Press, 1992.

[14] O. Khatib, “The potential field approach and operational spaceformulation in robot control,” in Adaptive and Learning Systems,pp. 367–377, Springer, 1986.

[15] J. Barraquand, L. Kavraki, R. Motwani, J.-C. Latombe, T.-Y. Li,and P. Raghavan, “A random sampling scheme for path planning,”





in Robotics Research (G. Giralt and G. Hirzinger, eds.),pp. 249–264, Springer London, 1996.

[16] R. Geraerts and M. H. Overmars, “A comparative study ofprobabilistic roadmap planners,” in Algorithmic Foundations ofRobotics V, pp. 43–57, Springer, 2004.

[17] K. Nishiwaki, J. Kuffner, S. Kagami, M. Inaba, and H. Inoue, “Theexperimental humanoid robot h7: a research platform forautonomous behaviour,” Philosophical Transactions of the RoyalSociety of London A: Mathematical, Physical and EngineeringSciences, vol. 365, no. 1850, pp. 79–107, 2007.

[18] R. Brooks, “A robust layered control system for a mobile robot,”Robotics and Automation, IEEE Journal of, vol. 2, pp. 14–23, Mar1986.

[19] M. J. Mataric, “Interaction and intelligent behavior.,” tech. rep.,DTIC Document, 1994.





[20] M. P. Georgeff and A. L. Lansky, “Reactive reasoning andplanning.,” in AAAI, vol. 87, pp. 677–682, 1987.

[21] J. Zhang and A. Knoll, Integrating Deliberative and ReactiveStrategies via Fuzzy Modular Control, pp. 367–385.Heidelberg: Physica-Verlag HD, 2001.

[22] J. S. Albus, “The nist real-time control system (rcs): an approachto intelligent systems research,” Journal of Experimental &Theoretical Artificial Intelligence, vol. 9, no. 2-3, pp. 157–174, 1997.

[23] A. Meystel, “Nested hierarchical control,” 1993.

[24] T. Fukuda and T. Shibata, “Hierarchical intelligent control forrobotic motion by using fuzzy, artificial intelligence, and neuralnetwork,” in Neural Networks, 1992. IJCNN., International JointConference on, vol. 1, pp. 269–274 vol.1, Jun 1992.





[25] R. C. Arkin and T. Balch, “Aura: principles and practice in review,”Journal of Experimental & Theoretical Artificial Intelligence, vol. 9,no. 2-3, pp. 175–189, 1997.

[26] E. Gat, “Integrating reaction and planning in a heterogeneousasynchronous architecture for mobile robot navigation,” ACMSIGART Bulletin, vol. 2, no. 4, pp. 70–74, 1991.

[27] L. Einig, Hierarchical Plan Generation and Selection for ShortestPlans based on Experienced Execution Duration.Master thesis, Universität Hamburg, 2015.


minfaculty departmentofinformatics - introductiontorobotics€¦ · minfaculty...

Documents