A Multi-Path Compilation Approach to Contingent Planning

We describe a new sound and complete method forcompiling contingent planning problems with sensingactions into classical planning. Our encodesconditional plans within linear, classicalplan. This allows our planner, MPSR, to reasonabout multiple future outcomes of sensing actions,and makes it less susceptible dead-ends. MPRS,however, generates very large planningproblems. To overcome this, we use an incompletevariant the method, based on state sampling,within online replanner. On most currentdomains, MPSR finds faster, although itsplans are often longer. But challengingvariant Wumpus dead-ends, smallerplans, scales much better.