Autonomous system for cross-country navigation

Autonomous cross-country navigation is essential for outdoor robots moving about in unstructured environments. Most existing systems use range sensors to determine the shape of the terrain, plan a trajectory that avoids obstacles, and then drive the trajectory. Performance has been limited by the range and accuracy of sensors, insufficient vehicle-terrain interaction models, and the availability of high-speed computers. As these elements improve, higher- speed navigation on rougher terrain becomes possible. We have developed a software system for autonomous navigation that provides for greater capability. The perception system supports a large braking distance by fusing multiple range images to build a map of the terrain in front of the vehicle. The system identifies range shadows and interpolates undersamples regions to account for rough terrain effects. The motion planner reduces computational complexity by investigating a minimum number of trajectories. Speeds along the trajectory are set to provide for dynamic stability. The entire system was tested in simulation, and a subset of the capability was demonstrated on a real vehicle. Results to date include a continuous 5.1 kilometer run across moderate terrain with obstacles. This paper begins with the applications, prior work, limitations, and current paradigms for autonomous cross-country navigation, and then describes our contribution to the area.