Efficient Planning with State Trajectory Constraints

This paper introduces a general planning approach with state trajectory constraints, a language extension that has recently been introduced to PDDL for running the Fifth International Planning Competition in 2006. In our approach PDDL inputs with state trajectory constraints are translated into ordinary (fully instantiated) PDDL. It exploits the automata-based interpretation of the constraints, and refers to planning via explicit-state model checking. We explain how trajectory constraints are translated into LTL and non-deterministic Büchi automata and how these automata are compiled back to ordinary PDDL predicates and actions. We provide promising initial experimental results with heuristic search planning.