Substructure discovery of macro-operators

A description is given of a system, PLAND, for discovering substructures in observed action sequences. The goal is to show how a system can learn useful macro-operators by observing a task being performed. An intelligent robot using this system could learn how to perform tasks by watching someone else, even if it does not possess a complete understanding of the actions being observed. Macro-operators are discovered within a specific context that provide the types of generalizations allowed in the discovery process and use the previously proposed macro-operators to build new ones. Background knowledge is used to determine which generalizations are appropriate and to control search. The system can discover syntactic structures (grammars) without background knowledge, but more meaningful and useful structures are discovered when background knowledge is incorporated into the process. The foundations of PLAND are in similarity-difference-based (SDBL) learning systems that perform conceptual clustering; however, unlike most SDBL systems, a large amount of background knowledge can be incorporated to improve learning effectiveness.<
>