Conformant Planning via Heuristic Forward Search

Conformant planning is the task of generating plans given uncertainty about the initial state and action effects, and without any sensing capabilities during plan execution. The plan should be successful regardless of which particular initial state we are in. This paper is motivated by the observation that (1) Conformant planning without conditional effects can be handled easily by any forward search planner; (2) To handle conditional effects, one can reason about the set of known facts following each sequence of actions, rather than explicitly enumerating the sets of possible worlds. Using this computation of known facts, we extend the classical planning system FF to the conformant setting. Our experimental evaluation shows Conformant-FF to be superior to the state-of-theart conformant planner MBP in a variety of benchmark domains.