Planning with Derived Predicates through Rule-Action Graphs and Relaxed-Plan Heuristics

The ability to express “derived predicates” in the formalization of a planning domain is both practically and theoretically important. The recent PDDL2.2 language supports derived predicates, which can be expressed by “domain rules”. We propose an approach to planning with derived predicates where the search space consists of particular graphs of actions and rules, called rule-action graphs, representing partial plans. We present (i) some techniques for managing domain rules in the context of a local search process for rule-action graphs, (ii) new heuristics for guiding the search, and (iii) a new method for restricting the search neighborhood to speed up the search. The proposed approach and techniques are implemented in a new version of the LPG planner, which took part in the fourth International Planning Competition showing good performance in many benchmark problems.