Learning for Planning

This example uses a planner in a simple domain. Initially there are no operators but a set of actions with unknown preconditions and effects. Over time the system will learn operators from observerd state transitions which will allow it to hopefully reach the goal.

• Loop:

  1. Try to find a plan

     • No plan found: try random action

     • Plan found: execute plan

  2. Record state transition data

  3. Learn a decision tree to predict action effects given action and state

  4. Convert decision tree to operators

Go here for instructions to run this example.