Fast Forward Planning System Based On Delayed Partly Reasoning

Artificial intelligence planning is an important research area in artificial intelligence andapplying heuristic search in planning has become a popular trend.Fast-Forward planning system (FF) is a successful state-space search planner, whichobtains heuristics via a relaxed planning graph to guide the enforced hill-climbing searchstrategy, has shown excellent performance in most STRIPS domains. When it comes to ADLdomains, FF handles actions with conditional effects in a way similar to factored expansion.The result is that enforced hill-climbing guided by the relaxed Graphplan always fails insome ADL domains. We have discovered that the reason behind this issue is the relaxedGraphplan's inability to handle relationships between actions' components. We propose anovel approach called delayed partly reasoning on a na?ve conditional-effects planning graph(DPR-NCEPG). We do not ignore actions' delete effects and consider restricted inducedcomponent mutual exclusions between factored expanded actions. Preliminary results showthat enforced hill-climbing while guided by DPR-NCEPG gains obvious improvements inmost ADL domains in terms of both solution length and runtime.While current methods for extracting heuristics are mostly based on fully ignoringactions' delete effects, this work shows that the efficiency of heuristic functions can beimproved by even reasoning on a part of actions' delete effects. Further, this work is helpfulfor improving the ability of planners that do planning with heuristic state space search whileexploiting the power of pruning on ADL domains.