The Footprint Principle for Heuristics for Probabilistic Planners

Probabilistic back-chaining planners, which use probabilities to represent and reason about uncertainty in the planning domain, typically have a larger search space than their classical counterparts. Therefore heuristics that can reduce their search effectively are even more important. The “footprint” principle leads to a family of heuristics for probabilistic planners produced by attempting to make subsequent refinements to a plan apply to disjoint sets of planning cases. Heuristics derived by this principle are shown to be effective for two probabilistic planners, Buridan and Weaver, which are organised around quite different search techniques. Probabilistic planners are needed that can use more compact representations of uncertainty than those that currently exist, and these planners will depend even more on the footprint principle and others like it.