Repairing Planning Domains Based on Lifted Test Plans

Knowledge engineering for AI planning remains a significant challenge, particularly in the creation and maintenance of accurate domain models. A recent approach to correcting flawed models involves using test plans: non-solution plans that are intended to be solutions. However, these plans must be grounded, which restricts the modeler's ability to specify repairs at various levels of abstraction, especially when only partial information about the grounding is available. In this paper, we propose a novel approach that extends domain repair capabilities to handle lifted test plans, in which action parameters can remain unspecified. We introduce a novel search algorithm along with a heuristic function for solving the problem with lifted test plans. Our experimental results demonstrate that the proposed approach efficiently solves a wide range of problems and finds close approximations to optimal solutions in the majority of cases.