Non-classical Planning with a Classical Planner: The Power of Transformations

Planning is the model-based approach to autonomous behavior where a predictive model of actions and sensors is used to generate the behavior for achieving given goals. The main challenges in planning are computational as all models, whether featuring uncertainty and feedback or not, are intractable in the worst case when represented in compact form. Classical planning refers to the simplest form of planning where goals are to be achieved by applying deterministic actions to a fully known initial situation. In this invited paper, I review the inferences performed by classical planners that enable them to deal with large problems, and the transformations that have been developed for using these planners to deal with non-classical features such as soft goals, hidden goals to be recognized, planning with incomplete information and sensing, and multiagent nested beliefs.