Using the Relaxed Plan Heuristic to Select Goals in Oversubscription Planning Problems

Oversubscription planning (OSP) appears in many real problems where finding a plan achieving all goals is infeasible. The objective is to find a feasible plan reaching a goal subset while maximizing some measure of utility. In this paper, we present a new technique to select goals "a priori" for problems in which a cost bound prevents all the goals from being achieved. It uses estimations of distances between goals, which are computed using relaxed plans. Using these distances, a search in the space of subsets of goals is performed, yielding a new set of goals to plan for. A revised planning problem can be created and solved, taking into account only the selected goals. We present experiments in six different domains with good results.