A Possibilistic Planner that Deals with Non-Determinism and Contingency

This paper proposes a new planning approach that authorizes an autonomous robot to reason about the inaccuracy of the world description and of its possible evolutions. We represent the uncertainty with the possibility theory; this allows us to distinguish between two types of nondeterminism: a non-determinism from insufficient modeling and a non-determinism from uncertainty. Besides, we introduce perception actions as well as a model of the environment dynamics through "contingent events". Finally, we present an implemented experimental planner, based on Graphplan search paradigm. This planner is able to produce plans that are robust with respect to contingent events, and whose goal-achieving ability is evaluated a priori. The obtained plans can be conformant or conditional depending on the context and the user requirements.