Presentation is loading. Please wait.

Presentation is loading. Please wait.

Introduction to SOAR Based on “a gentle introduction to soar: an Architecture for Human Cognition” by Jill Fain Lehman, John Laird, Paul Rosenbloom. Presented.

Similar presentations


Presentation on theme: "Introduction to SOAR Based on “a gentle introduction to soar: an Architecture for Human Cognition” by Jill Fain Lehman, John Laird, Paul Rosenbloom. Presented."— Presentation transcript:

1 Introduction to SOAR Based on “a gentle introduction to soar: an Architecture for Human Cognition” by Jill Fain Lehman, John Laird, Paul Rosenbloom. Presented by Roman Ilin

2 Unified Theories of Cognition 1980, Newell started the project Task: to find a set of computationally realizable mechanisms and structures that can answer all the questions about cognitive behavior

3 Architecture Analogy with computer hardware (fixed) – software (changeable) BEHAVIOR = ARCHITECTURE + CONTEXT Architecture reflects designer’s assumptions about the context. In general, Architecture is a theory of what is common among much of the behavior at the level above it. Cognitive Architecture is a theory of the fixed mechanisms and structures that underlie human cognition. SOAR is a cognitive architecture

4 What cognitive behaviors are common? 1.Goal oriented 2.Reflects a rich, complex, detailed environment 3.Requires a large amount of knowledge 4.Requires use if symbols and abstractions 5.Flexible and a function of the environment (real time) 6.Requires learning from the environment and experience

5 CONTEXT

6 BEHAVIOR = ARCHITECTURE + CONTEXT CONTEXT is a theory about the knowledge the agent has that contributes to the behavior

7 Example of knowledge categories

8 Behavior as Movement through Problem Spaces

9 Formalize problem space – goal, states and operators, and the principle of rationality

10 Connecting Content (knowledge) to Architecture Need Domain Independent Level of knowledge description It is “Goal Context” – a set of four (kinds of) things. {goals, problem spaces, states, operators} Knowledge is represented in terms of the above four things

11 Goal Context Note, Single structure can be used for both “acting” and “thinking about acting”

12 Memory, Perception, action and Cognition Long Term Memory (LTM) – knowledge that is independent of the current goal Working Memory (WM) – current occurrence of some portion of that knowledge Decision Cycle – to tie LTM to WM

13 LTM – if – then statements

14 Decision Cycle – two phases Elaboration –Contents of WM are matched against the IF parts of LTM Decision –Select of the suggested operators

15

16

17 What if decision cannot be made? Impasse results in switching the problem space SOAR defines fixed set of domain independent impasses. –Resolve tie impasse –Fail to decide impasse –…

18

19 Finding Solution using sub goal two more operators –Augment –Evaluate

20 Augment

21 Evaluation Operator

22

23 LEARNING, FINALLY Practice improves what we do Since behavior = architecture + content And architecture is fixed Content must change (learn) Do it by adding new entries in LTM Creates a “Chunk” by using parts of the environment existing in pre-impasse environment that were used to achieve the result Chunking is deductive learning

24 Putting it all together

25 ROBO SOAR: An Integration of external interaction, planning. And learning using Soar. John. E. Laird, Eric. S. Yager, M. Hucka, C. M. Tuck, 1991 Presented by Roman Ilin

26 THE ROBOT AND ITS TASKS

27 Capabilities Problem Solving with Incomplete Information Problem solving with delayed perception Planning Learning from external guidance Interruption and reactivity Improve efficiency Improve correctness

28 System Architecture

29 Primitive Operators – commands sent to the robot controller (snap-in, snap-out not shown)

30 Initial Operators ALIGN-BLOCKS TURN-OUT-LIGHT Light has a preference Initially the operators will lead to impasses and learning.

31 Example of problem solving goals: align-blocks, align block-pair, puma-arm- command

32 Guided Problem Solving – planning I depth first guided search.

33 Guided Problem Solving – planning I chunking

34 Refining Knowledge I triangular blocks

35 Refining Knowledge II triangular blocks


Download ppt "Introduction to SOAR Based on “a gentle introduction to soar: an Architecture for Human Cognition” by Jill Fain Lehman, John Laird, Paul Rosenbloom. Presented."

Similar presentations


Ads by Google