Abstraction Learning in Prodigy

Abstraction Learning in Prodigy

ALPINE produces abstraction hierarchies to help control the problem solver's search. To do this, the module employs an algorithm that takes a set of operators and a problem domain as input then partitions the literals in the operators' preconditions, addlists, and deletelists to form a hierarchy of abstraction spaces. The literals in a given level of the hierarchy are constained to be achievable without the addition or deletion of a literal at a higher level in the hierarchy. The higher levels represent more abstract descriptions of the domain. The lower levels represent the domain in more detail. The problem solver can plan at a high level of abstraction. The most important preconditions and effects are considered in planning at this level. The plan is then refined as needed by planning at lower levels.


Return to the top of this architecture.

Go to a discussion of this capability for multiple architectures.


Current Location: Prodigy-Capabilities-Multi-Method Learning-Abstraction Learning

Go to NEXT page.