Metareasoning and Path Planning for Autonomous Indoor Navigation

Susan Epstein, Raj Korpan

Abstract: A planner for indoor navigation in an unknown space should anticipate the perils of sensor and actuator errors. The approach described here uses a spatial model whose elements generalize continuous affordances from discrete data to support robust hierarchical planning. A novel reactive controller intervenes to address lack of progress toward a target and to improve plans opportunistically, during their execution. A metareasoning architecture seamlessly integrates these components as hierarchical decision making. Empirical results demonstrate the flexibility of this approach and the role of the spatial model in three challenging real worlds.