Finding and combining goal interaction

Most of existing planning methods hitherto assume that each goal is achievable and goals are independent, which is very impractical in our daily life. In fact there often exist many positive and negative interactions between goals, and the orders suggested by the dependence between goals can be further used as heuristic information in planning. Although recently Over-Subscription Planning problems (OSP) received more and more attention in planning community, few methods have been proposed to find out these interactions before planning. So in this paper, we introduce a method to find out primary interactions between atomic goals and orders can be inferred from these primary interactions. We define six types of interactions between goals and combination operations for them, which can be regarded as a groupoid algebra system. The orders suggest by interactions can be used as heuristic for planner. In practical, our framework can be a preprocessor of any classical planner for OSP handling, and we combine our suggested orders with other goal ordering techniques to build a more effective heuristic strategy. Besides tackling OSP, the time of processing goal relations inference is negligible comparing to the planning time, so we believe this method can serve as a new heuristic technique. We provide preliminary empirical results in a planner we called FF^ps which demonstrate the effectiveness.