Robust Mobile Robot Navigation using Partially-Observable Semi-Markov Decision Processes

Mobile robots operating in dynamic ooce environments need robust navigation algorithms capable of dealing with uncertainties arising from sensor and actuator errors. Such algorithms also need to be able to synthesize plans that take into account temporal aspects of their environment, such as crowded corridors, and busy intersections. This paper describes a novel mobile robot navigation architecture based on Partially-Observable Semi-Markov Decision Processes (POSMDP). This model extends previous work on POMDP-based navigation architectures, which assumed a discrete-time temporal model of actions. We show that the POSMDP model allows the robot to explicitly model the transition time of diierent actions in various locations in the environment (e.g. passing through intersections takes more time than going through corridors). The paper describes experimental results of using the POSMDP architecture on a real mobile robot called PAVLOV. The results show that the robot is able to traverse paths of several hundred meters and also plan to avoid crowded corridors.