Abstract:
This work presents a planning paradigm which integrates the flexibility and clear semantics of decision theory with search procedures providing many of the efficiency properties of classical heuristic planning methods. We define plans in a manner similar to the classical paradigm but extended to represent nondeterministic actions, using utilities to represent goals. We then outline a search procedure for good plans which is guided by decision theory, and discuss knowledge representation issues from the point of view of improving decisions. This paradigm puts plan choice on the same level as action choice, and has as emergent properties variants of desirable features of more classical planners, such as abstraction and reactivity.