Pruning bad quality causal links in sequential satisfying planning

Although current sequential satisficing planners are able to find solutions for a wide range of problems, the generation of good quality plans still remains a challenge. Anytime planners, which use the cost of the last plan found to prune the next search episodes, have shown useful to improve the quality of the solutions. With this in mind this paper proposes a method that exploits the solutions found by an anytime planner to improve the quality of the subsequent ones. The method extracts a set of causal links from the first plans, the plans with worse quality, and creates a more constrained definition of the planning task that rejects the creation of these causal links. The performance of the proposed method is evaluated in domains in which optimization is particularly challenging.