Reasoning about partially observed actions

摘要: Partially observed actions are observations of action executions in which we uncertain about the identity objects, agents, or locations involved (e.g., know that move(?o, ?x, ?y) occurred, but do not ?o, ?y). Observed-Action Reasoning is problem reasoning world state after a sequence partial and states. In this paper formalize Reasoning, prove intractability results for current techniques, find tractable algorithms STRIPS other actions. Our new update representation all possible states (the belief state) logic using logical constants unknown objects. A straightforward application idea incorrect, identify add two key amendments. We also present successful experimental our algorithm Blocks-world domains varying sizes Kriegspiel (partially observable chess). These promising relating sensors with symbols, partial-knowledge games, multi-agent decision making, AI planning.