Pruning Search Space for Heuristic Planning through Action Utility Analysis

This paper studies how to prune search space under the Relaxed Planning Graph heuristic framework which is firstly proposed in FF. Based on the observation of relations between action and proposition layers, we present a new and high-quality domain-independent pruning strategy for forward-chaining heuristic planning through Action Utility Analysis. The proposed strategy extracts so-called directly-used actions from relaxed planning graph as promising successors. Its pruning result is in general better than the helpful actions strategy. Our strategy can be used in any progression state space search framework. Experiments in the STRIPS benchmarks of International Planning Competitions (IPC) show that our pruning strategy along with algorithms decreases the search space effectively, and can outperform helpful action strategy of FF.