This project contains a version of the water jug problem that is formulated for hierarchical task decomposition. It involves two levels of problem spaces. The top level has three operators: fill, empty or pour. The next level consists of three operators: pick-up, fill-jug and put-down and they arise in an operator no-change for the super state.

Note: This agent works with chunking, which compiles the actions in the substates into rules that apply at the top-state. Use the command "learn -e" to turn on chunking, run the agent once, init-soar then run the agent again to see how the learned productions eliminate the need to subgoal in the second run.

Soar capabilities
  • Hierarchical task composition via subgoaling
  • Can learn procedural knowledge (enable with 'learn always')
Download LinksExternal Environment
  • None.
Default Rules
  • None.
Associated Publications
  • None.
Developer
  • John Laird
Soar Versions
  • Soar 8,9
Project Type
  • VisualSoar