Partially Observable Markov Decision Process models (POMDPs) have been applied to low-level robot control. We show how to use POMDPs differently, namely for sensor-planning in the context of behavior-based robot systems. This is possible because solutions of POMDPs can be expressed as policy graphs, which are similar to the finite state automata that behavior-based systems use to sequence their behaviors. An advantage of our system over previous POMDP navigation systems is that it is able to find close-to-optimal plans since it plans at a higher level and thus with smaller state spaces. An advantage of our system over behavior-based systems that need to get programmed by their users is that it can optimize plans during missions and thus deal robustly with probabilistic models that are initially inaccurate.
展开▼
