intelligent robotics

21
Intelligent Robotics Jeremy Wyatt Thanks to: Nick Hawes, Aaron Sloman, Michael Zillich, Somboon Hongeng, Marek Kopicki

Upload: carol

Post on 19-Mar-2016

55 views

Category:

Documents


1 download

DESCRIPTION

Intelligent Robotics. Jeremy Wyatt Thanks to: Nick Hawes, Aaron Sloman, Michael Zillich, Somboon Hongeng, Marek Kopicki. The Whole Iguana. AI commonly studies aspects of intelligence separately: narrow domain high performance - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Intelligent Robotics

Intelligent Robotics

Jeremy Wyatt

Thanks to: Nick Hawes, Aaron Sloman, Michael Zillich, Somboon Hongeng, Marek

Kopicki

Page 2: Intelligent Robotics

The Whole Iguana AI commonly studies aspects of intelligence separately:

narrow domain high performance

In 1976, philosopher Dan Dennett suggested putting it all together, but with a low level of performance

In fact people had been trying to build integrated systems for some twenty years by then

Page 3: Intelligent Robotics

Shakey the robot 1970 - Shakey the robot reasons about

its blocksBuilt at Stanford Research Institute, Shakey was remote controlled by a large computer. It hosted a clever reasoning program fed very selective spatial data, derived from weak edge-based processing of camera and laser range measurements. On a very good day it could formulate and execute, over a period of hours, plans involving moving from place to place and pushing blocks to achieve a goal.

(Hans Moravec)

Page 4: Intelligent Robotics

Shakey: key ingredients World model used logical representations

type(r1,room)in(shakey,r1)in(o1,r2)type(d1 door)type(o1 object)type(f3 face)type(shakey)at(o1 15.1 21.0)joinsfaces(d2 f3 f4)joinsrooms(d2 r3 r2)…

shakey

30

20

10

0

0 10 20

r3

f4 f3

d2

d1

f2

f1

r1

r2o1

Page 5: Intelligent Robotics

Planning Shakey used a form of planning called goal regression

Idea: find an action that directly achieves your goal, and then actions to achieve the first action’s preconditions, etc…

e.g. Blocked(d1,X)

Let’s see Shakey solve a problem

block_door(D,Y)preconditions: in(shakey,X) & in(Y,X)

& clear(D) & door(D)& object(Y)

delete list: clear(D)add list: blocked(D,Y)

shakey

30

20

10

0

0 10 20

r3

f4 f3

d2

d1

f2

f1

r1

r2o1

Page 6: Intelligent Robotics

Lessons from nature Gannets – wings half open to

control dive

Fold wings to avoid damage

Not at a constant distance, but ata constant time

Birds have detectors that calculate time to impact

Page 7: Intelligent Robotics

Lessons from nature All naturally occuring intelligence is embodied

So robots are in some ways similar systems

Robots, like animals exploit their environments to solve specific tasks

“There are no general purpose animals … why should there be general purpose robots?”

David MacFarland

Page 8: Intelligent Robotics

Behaviour Based Robots Inspired by simpler creatures than

humans

Throw away most representations

Throw away most reasoning

Build your robot out of task specific behaviours

Page 9: Intelligent Robotics

Pushing the behaviour based envelope

Behaviour based systems can display quite sophisticated behaviour, particularly for interaction

But they don’t have understanding because they don’t have representations

Page 10: Intelligent Robotics

The age of data In the 1990s people were finally beginning to have success

with representation driven approaches

One key has been the use of probabilistic methods

These are data intensive and require very strong assumptions about the learning task

Stanley

Page 11: Intelligent Robotics

Robots that understand

Internal structure to represent the meaning of the utterance

e.g. “The orange ball”

B1:phys-object ^ ball<property> C1:colour ^ orange

Page 12: Intelligent Robotics

Learning object appearances

Page 13: Intelligent Robotics

Learning names and appearances of objects

Page 14: Intelligent Robotics

Cognition requires attention

Object recognition is unreliable and expensive

We can use bottom up salience to make it more efficient

Page 15: Intelligent Robotics

Salience can be modulated by languageDirecting processing of the visual scene

Page 16: Intelligent Robotics

The Whole Iguana: coming full circle Collection of loosely coupled sub-

architectures

Each sub-architecture contains processing elements that update structures within a working memory

WM are typically only locally read & write (bar privileged sub-architectures)

Processing controlled by local and global goals and managers

Knowledge management by local and global ontologies

SensorActuator

ProcessorWorking

Memory

Manager

Page 17: Intelligent Robotics

Movie goes here

Illustration: Cross Modal Ontology Learning Architectures

Page 18: Intelligent Robotics

Linguistically Driven ManipulationIllustration: Language Driven Manipulation Architectures

Goals are raised by language

Reference to objects by learned features

Robot plans intentional actions using logical planner

Intention shifting is handled via execution monitoring and continual replanning

Communication SA

Communication SA

Communication SA

Binding SA

Communication SA

Visual SA

Communication SA

Spatiotemporal SA

Communication SA

Coordinator SACommunicatio

n SAPlanning SA

Communication SA

Manipulation SA

Page 19: Intelligent Robotics

ROI 1

SO 1

ROI 2

SO 2

Inst 1

Qual Spatial Relations

Inst 2

Object locations

Communication SA

Binding SA

Visual SA

“Put the blue thing to the right of the red thing”

Parse + Dialogue

Interpretation

Coordination SA

Spatial SA

Raise Planning Goal

Goal LF

Planning SA

Object locations

Qual Spatial Relations

Object locations

Qual Spatial Relations

ROI 1

SO 1

Inst 1

ROI 2

SO 2

Inst 2

Manipulation SA

MAPL Goal Plan Plan Step

Vis Servo

Manip Goal Executed

Execution Check

ROI 1

SO 1

ROI 2

SO 2

Raise Manip Goal

Inst 1 Inst 2

Page 20: Intelligent Robotics

Movie goes here

Illustration: Language Driven Manipulation Architectures

Page 21: Intelligent Robotics

Wrap up Robotics gets to the heart of big issues in AI

There are enormous engineering and scientific challenges

There is a tension between different approaches:• Representation heavy, top-down approaches to cognition• Representation light, bottom approaches

The fun is in linking these