ViSoar A Graphical Agent Development Environment for Soar

1
ViSoar - A Graphical Agent Development
Environment for Soar

• Tony Hirst
• ICSRG

2
Overview

• What is Soar?
• What is ViSoar?
• Why Bother?
• Demo

3
What is Soar?

• A production system architecture
• Developed as a cognitive modelling architecture
• BUT
• Soar may be also used as an agent architecture
  for the construction of psychologically
  plausible autonomous agents

4
Knowledge in Soar Agents

• Architectural constraints on representation
• lead to idea of Idealised Soar Model (ISM)
• Explicit scaffolding knowledge
• production names
• file structure for larger rulesets
• Implicit knowledge
• operator/problem-space hierarchy
• avtree ontology

5
Typing Soar Productions I

• State elaboration
• default/context defining elaborations
• contingent attributes and values
• Productions for making permanent changes to state
  elements
• operator proposition, search control,
  application, (termination)

6
Typing Soar Productions II

• Productions for interfacing with an external
  environment
• input-link
• receiver information about the environmental
  state
• receive messages from other agents
• output-link
• perform actions on the environment
• transmit messages to other agents

7
Explicit Scaffolding Knowledge

• File conventions (partitioned production storage)
• file contents
• filenames
• Production name conventions
• reflect problem-space/operator hierarchy
• reflect production type

8
Operator Problem-Space Hierarchies

• Operator hierarchy
• (ltsgt superstate (ltssgt operator.name parent)
• (ltsgt operator.name child)
• Problem-space hierarchy
• explicit
• (ltsgt problem-space.name whatever)
• (ltsgt superstate.operator.name whatever)
• implicit

9
The avtree I - Implicit vs. Explicit Problem
Spaces

• Explicit problem-spaces are identified with the
  problem-space.name attribute
• Implictly defined problem-spaces reflect the
  state elements manipulated by productions acting
  within the problem-space
• The avtree for productions whose behaviour we
  associate with activity in a given problem-space
  defines the problem-space

10
The avtree II - Example

• (state ltsgt lev1a (ltlev1agt lev2 ltlev2gt)
  lev1b val1)
• (ltlev2gt lev3a ltlt val3a val3b gtgt -lev3b ltxgt)
• --gt
• (ltlev2gt lev3b val3c )
• results in the avtree
• lev1a
• lev2
• lev3a
• val3a, val3b
• lev3b
• val3c
• lev1b
• val1

11
The avtree III - Element Types

• The avtree for a ruleset essentially provides a
  hierarchically structured ontology
• major attributes (have attributes as values)
• minor attributes (have constant values)
• (fixed) values
• single attributes
• multi-attributes

12
ViSoar

• ViSoar - an integrated visual development and
  debugging environment for the Soar cognitive
  agent architecture.

13
Available Tools

• Coding Tools
• Soar Development Environment (SDE)
• emacs addon developed for Soar 6
• Debugging Tools
• Soar architecture commands
• GUI front end in form of Tcl-Soar Interface (TSI)
• Development Tools
• Task AcQuisition Language (TAQL)
• Visual Soar

14
Philosophy

• Visual environment
• Written in Tcl/Tk
• Exploit Soar architecture at command rather than
  code level
• Facilitate the construction of typo free Soar code

15
What does ViSoar provide?

• An integrated visual Soar development environment
• code generation tools (deViSoar)
• offline reverse engineering tools (reViSoar)
• high level structured editor and reverse
  engineering tool (Phoebe)

16
deViSoar

• Text editor
• avtree ontology editor
• Automatically generated skeleton productions
• Various Soar code manipulating tools
• OR
• NOT

17
reViSoar

• Reverse engineering of legacy Soar agents
• Utilises an offline operator hierarchy analyser
  (OpHelia)
• extracts implicit structure of rulesets
• operator/problem space hierarchy
• productions related to particular
  operators/problem spaces
• Generates avtree that may be passed to deViSoar

18
Phoebe

• High level, structured editor for building task
  hierarchies using Idealised Soar Model (ISM)
• Automatic code generation and structured
  code-save capabilities
• Automated reverse engineering of Soar code into
  ISM format
• Awaiting full integration with deViSoar and
  reViSoar