Repair Suggestions for Planning Domains with Missing Actions Effects

Alba Gragera, Raquel Fuentetaja, Ángel García-Olaya, Fernando Fernández

Keywords: Classical Planning, Incomplete Domains

TL;DR: Repair Suggestions for Planning Domains with Missing Actions Effects

Abstract: In Automated Planning (AP) a proper definition of the domain and problem files is assumed. However, producing complete model descriptions is a time consuming and challenging task, especially for non-experts. It is easy to make mistakes when creating formal models, turning the planning task unsolvable for the planners. This can happen if the initial state is not fully and properly specified, some actions are missing, or some actions are incomplete. Explaining the absence of solution in such cases is essential to help humans in the development of AP tasks. In this paper we focus on repairing planning models where the effects of some actions are incomplete. We propose a compilation of the unsolvable task to a new extended planning task, where the actions are allowed to insert possible missing predicates in their effects. The solution to such task is a plan that achieves the goals of the original problem while also warning about the modifications that were necessary to do so. Experimental results show this approach can be effectively used to repair incomplete planning tasks across different planning domains.