Graphplan Based Conformant Planning with Limited Quantification

Conformant planners solve problems with a correct but incomplete description of the initial state, by finding plans that are valid for all possible assignments of the unknown atoms. Most conformant planners, however, do not handle universal quantification, which is a problem when the set of all domain objects is unknown or very large, and thus can not be enumerated. This paper introduces PSIGRAPH, a conformant planner that operates with universally quantified statements in the initial and goal states, as well as in the action preconditions. Thus, PSIGRAPH does not need to know the complete set of domain objects. We present the algorithm and the results of its experimental evaluation, showing that PSIGRAPH is competitive with other conformant planners. PSIGRAPH is based on Graphplan, but differs from previous approaches such as Conformant Graphplan in that it does not create multiple plan graphs