"LPRPG-P: Relaxed Plan Heuristics for Planning with Preferences." A. J. Coles and A. I. Coles. Proceedings of the Twenty First International Conference on Automated Planning and Scheduling (ICAPS-11). June 2011.
Abstract: In this paper we present a planner, LPRPG-P, capable of reasoning with the non-temporal subset of PDDL3 preferences. Our focus is on computation of relaxed plan based heuristics that effectively guide a planner towards good solutions satisfying preferences. We build on the planner LPRPG, a hybrid relaxed planning graph RPGP--linear programming (LP) approach. We make extensions to the RPG to reason with propositional preferences, and to the LP to reason with numeric preferences. LPRPG-P is the first planner with direct guidance for numeric preference satisfaction, exploiting the strong numeric reasoning of the LP. We introduce an anytime search approach for use with our new heuristic, and present results showing that LPRPG-P extends the state of the art in domain-independent planning with preferences.