Abstract:

Based on Graphplanencoding, a new encoding called PMAbased encoding by reducing the mutex axioms of actions, is proposed. The soundness and completeness of the new encoding are justified. The preprocessing method for goal states during encoding solving process is also proposed. The corresponding planner called PMASP based on the above encoding is designed and implemented, and test with benchmarks adopted by International Planning Competition. Comparing with the planner SATPLAN2006, the experimental results demonstrate that, for both tested domains, the scales of encodings of PMASP are compacted and the solver efficiencies are improved. For sequential planning problem Blocks World, the scales of encodings of PMSSP are compacted nearly 40% and solver efficiencies are increased nearly two times. For parallel planning problem Logistics, the scales of encodings of PMASP are compacted more than 80%.

Key words: artificial intelligence, planning as satisfiability, Graphplanbased encoding, mutex axiom, planner