Planning in the Entropy Reduction Engine

The integration of planning and scheduling is one of the major goals of the design of ERE. Planning, as it were, is done exclusively by the Projector module. When given a strategy from the Reductor, it builts a graph of possible future behaviors and uses the behavior constraints to prune the tree of unacceptable behaviors. The Projector can also use situated control rules (also used for reactivity) to avoid considering some alternative actions.

Since the Projector takes probabilities into account, it can accomodate uncertain outcomes of actions (which would include failures) and generate conditional plans. This is advantageous in uncertain or unpredictable environments. Once plans are formed, they are compiled into situated control rules to be used by the Reactor.

The planner operates in an anytime fashion, which affords real-time performance in dynamic environments.

Thus ERE is continually shifting its behavior from planning-based to reaction-based. Planning is only invoked when there are no situated control rules that apply; that is, there is no immediate knowledge as to what to do. This combination of impasse-driven planning, and compilation of planning to reaction is similar in nature to Soar.

Return to the top of this architecture.

Go to a discussion of this capability for multiple architectures.

Current Location: Entropy Reduction Engine-Capabilities-Planning

Go to NEXT page.