RALPH

• Heterogeneous knowledge representation

  Six different types of knowledge are used by the system in making decisions, labeled A-F.
  • Type A knowledge specifies the probability of a specific state node given information about the status of other state nodes occuring during the same time period.

  • Type B knowledge specifies how an action will affect other nodes.

  • Type C knowledge specifies utility functions on the state.

  • Type D knowledge specifies production system-type condition-action rules.

  • Type E knowledge specifies the utility of condition-action pairs.

  • Type F knoweldge specifies the utility of future states.

• Meta-decisions

  Given suggested actions from all four Execution Architecture, a decision must be made on which of the four actions to take. Ralph contains a meta-level decision maker that, based on its knowledge about the performance of each EA given a certain amount of computation time, decides which EA will most likely suggest the best course of action.

• Meta-planning

  At each step of plan execution, Ralph compares the expected utility of executing the next step in the plan with the expected utility of replanning. If replanning ever becomes a more promising option, the uses the Plan Modification Focuser to find promising search areas for the Plan Modification module, which executes the replanning.

• Multiple Execution Architectures

  Ralph contains four different Execution Architectures, each using a different method to make decisions as to what action to exectute. These Execution Architectures each use different combinations of the six types of knowledge, resulting in alternate methods of decision making.