Plan Existence Verification as Symbolic Model Checking

Abstract

Automated Planning is the subarea of AI concerned with the generation of a plan of actions for an agent to achieve its goals. State-of-the-art planning algorithms are based on heuristic search. However, the inexistence of a plan can be a challenge for such planners, since they are not always able to discern the difficulty of finding a solution from its inexistence. The problem of plan existence verification, called planex, is computationally hard. Thus, in 2016, the planning community held for the first time the Unsolvability International Planning Competition (UIPC), which aims to evaluate algorithms on the task of verifying plan existence. The aim of this paper is to propose a new algorithm to solve the planex problem that is based on symbolic model checking approach. The proposed algorithm differs from others based on model checking in two points: (i) it is able to reason about the actions represented in PDDL (Planning Domain Description Language) and; (ii) it is based on the α-CTL logic, whose semantics takes into account the actions responsable for the state transitions. We also evaluate the proposed alorithm over the UIPC planning benchmark problems.