Symmetry Detection and Breaking in Cost-Optimal Numeric Planning

Alexander Shleyfman, Ryo Kuroiwa, Chris Beck

Keywords: numeric planning, symmetry breakin

TL;DR: adjusting problem description graph to detect symmetris in numeric planning

Abstract:

One of the main difficulties affecting the use of domain-independent numeric planning is the complexity of the search problem.
The exploitation of structural symmetries in a search problem can constitute an effective method of pruning search branches that may lead to exponential improvements in performance. Over a decade now, symmetry breaking techniques have been successfully used within both optimal and satisficing classical planning. In this work, we show that symmetry detection methods applied in classical planning with some effort can be modified to detect symmetries in linear numeric planning. The detected symmetry group, thereafter, can be used almost directly in the A*-based symmetry breaking algorithms such as DKS and Orbit Space Search. We empirically validate that symmetry pruning can yield a substantial reduction in the search effort, even if algorithms are equipped with a strong heuristic, such as LM-cut.