Tree-REX: SAT-Based Tree Exploration for Efficient and High-Quality HTN Planning

Abstract

In this paper, we propose a novel SAT-based planning approach to solve totally ordered hierarchical planning problems. Our approach called “Tree-like Reduction Exploration” (Tree-REX) makes two contributions: (1) it allows to rapidly solve hierarchical planning problems by making effective use of incremental SAT solving, and (2) it implements an anytime approach that gradually improves plan quality (makespan) as time resources are allotted. Incremental SAT solving is important as it reduces the encoding volume of planning problems, it builds on information obtained from previous search iterations and speeds up the search for plans. We show that Tree-REX outperforms state-of-the-art SAT-based HTN planning with respect to run times and plan quality on most of the considered IPC domains.