Deterministic planning in incompletely known domains with local effects

Abstract

Conformant planning has been traditionally studied in the form of classical planning extended with a mechanism for expressing unknown facts and/or disjunctive knowledge. Despite a sizable body of research, most approaches do not attempt to move beyond essentially propositional planning. We address this shortcoming by defining conformant planning in terms of the situation calculus semantics and use recent advances in the fields of first-order knowledge base progression and query answering to develop a sound and complete conformant planning algorithm capable of handling knowledge defined in an expressive fragment of first-order logic. We implement a prototype planner and evaluate its performance on several existing domains.