Unifying SAT-based and graph-based planning

The Blackbox planning system unifies the plan-ning as satisfiability framework (Kautz and Sel-man 1992, 1996) with the plan graph approach to STRIPS planning (Blum and Furst 1995). We show that STRIPS problems can be directly translated into SAT and efficiently solved using new random-ized systematic solvers. For certain computation-ally challenging benchmark problems this unified approach outperforms both SATPLAN and Graph-plan alone. We also demonstrate that polynomial-time SAT simplification algorithms applied to the encoded problem instances are a powerful com-plement to the “mutex ” propagation algorithm that works directly on the plan graph. 1