Action Usability via Deadend Detection

Abstract

The vast majority of planning problem models are incorrect, incomplete, or simply inconsistent. This is particularly an issue during the development process of a planning model. However, there are few existing debugging tools for modeling to aid with this issue. In this paper, we introduce a method for detecting actions that are deemed unusable, which may naturally be a result of modelling errors. We use action usability (or reachability) detection for on-the-fly diagnostics of planning models. In our technique, each action is checked for usability via problem reformulation and unsolvability detection. Through the analysis of usability, this technique could improve the modeling process directly, and we have demonstrated this capability through the tight integration with the online PDDL editor at Planning.Domains.

Publication
ICAPS Workshop on Knowledge Engineering for Planning and Scheduling (KEPS)
Date
Links