Mobility planning for autonomous navigation multiple robots in unstructured environments

In this paper we describe an approach to mobility planning for autonomous navigation of multiple vehicles unstructured environments. This approach is suitable for complex missions in which multiple goals with sequencing constraints may be specified. Ar the lowest level, this approach uses command arbitration in order to combine command recommendation from driving behaviors. Driving behaviors include local behaviors such as obstacle avoidance or goal tracking, and strategic behavior such as route planning to distant goals. Obstacle avoidance is performed by evaluating a Jived command set with respect to a terrain map built from 3-D sensors. Route planning is based on the D* dynamic route planner which evaluates the command set based on the currently optimal path to the next goal. The path is updated frequently based on map updates from the obstacle avoidance sensors. On top of this behavior-based system, an additional planner called GRAMMPS, continuously analyzes the outputs of the route planners running on each of the vehicles and re-computes optimally the allocation of goals to vehicles ill order to minimize time and distance travelled. This approach has be exercised extensively with two vehicles using stereo and ladar for obstacle detection and map building. Experiments were conducted over long distances with missions of up to nine intermediate goals in different configurations. We report of the implementation of this system and on its performance.