Assessing and Generating Robust Plans with Partial Domain Models

Most current planners assume complete domain models and focus on generating plans that are correct with respect to them. Unfortunately, assuming model completeness is unrealistic in the real world, where domain modeling remains a hard, labor-intensive and errorprone task. While domain experts cannot guarantee completeness, often they are able to circumscribe the incompleteness of the model by providing annotations as to which parts of the domain model may be incomplete. In such cases, the goal of planning would be to generate plans that are robust with respect to any known incompleteness of the domain. Doing this requires both a formal framework for assessing plan robustness and a methodology for guiding a planner’s search towards robust plans. In this paper we formalize the notion of plan robustness with respect to a partial domain model, show a way of reducing exact robustness assessment to model-counting, and describe methods of approximately assessing plan robustness. We propose a heuristic search approach using model-counting techniques on top of the FF planner to generate plans that are not only correct but also robust, and present experimental results showing the effectiveness of this approach.