Applying Marginal MAP Search to Probabilistic Conformant Planning: Initial Results

In this position paper, we present our current progress in applying marginal MAP algorithms for solving the conformant planning problems. Conformant planning problemis formulated as probabilistic inference in graphical models compiled from relational PPDDL domains. The translation from PPDDL into Dynamic BayesianNetwork is developed by mapping the SAT encoding of the ground PPDDL into factored representation. We experimented with recently developed AND/OR branchand bound search algorithms for marginal MAP over instances from the international planning competition domains, and we show that several domains were solved efficiently.