Compiling uncertainty away: solving conformant planning problems using a classical planner (sometimes)

• Authors: Héctor Palacios;Héctor Geffner
• Venue: AAAI'06 Proceedings of the 21st national conference on Artificial intelligence - Volume 1
• Year: 2006

Even under polynomial restrictions on plan length, conformant planning remains a very hard computational problem as plan verification itself can take exponential time. This heavy price cannot be avoided in general although in many cases conformant plans are verifiable efficiently by means of simple forms of disjunctive inference. This raises the question of whether it is possible to identify and use such forms of inference for developing an efficient but incomplete planner capable of solving non-trivial problems quickly. In this work, we show that this is possible by mapping conformant into classical problems that are then solved by an off-the-shelf classical planner. The formulation is sound as the classical plans obtained are all conformant, but it is incomplete as the inverse relation does not always hold. The translation accommodates 'reasoning by cases' by means of an 'split-protect-and-merge' strategy; namely, atoms L/Xi that represent conditional beliefs 'if Xi then L' are introduced in the classical encoding, that are combined by suitable actions to yield the literal L when the disjunction X1 ∨ ... ∨ Xn holds and certain invariants in the plan are verified. Empirical results over a wide variety of problems illustrate the power of the approach.