Planning with Sensing: Scaling Up

We develop a domain-independent action selection mechanism for planning with sensing that does not build full contingent plans but just chooses the action to do next in a closedloop fashion. For this to work, the mechanism must be fast and informed. We take advantage of recent ideas that allow delete and precondition-free contingent problems to be converted into conformant problems, and conformant problems into classical ones, for mapping the action selection problem in contingent planning to an action selection problem in classical planning. The formulation is then tested over standard benchmarks and over contingent problems with exponential size plans. The proposed action selection mechanism is shown to scale up in both settings, producing behaviors that lead to the goal in spite of the lack of explicit contingent plans.