Obstacles Fig. 8. Example of Rough Terrain Obstacle Avoidance Fig. 7. New Paths with Obstacle Avoided

Autonomous Cross-Country Navigation requires a system which can support a rapid traverse across challenging terrain while maintaining vehicle safety. This work describes a system for autonomous cross country navigation as implemented on the NavLab II, a computer-controlled off-road vehicle at Carnegie Mellon. The navigation software is discussed. The perception subsystem constructs digital maps of the terrain from range sensor data in real time. The planning subsystem uses a generate-and-test scheme to find safe trajectories through the map, and validates them through simulated driving. The planning process considers both kinematic and dynamic constraints on vehicle motion. The system was successful in achieving 300m autonomous runs on moderate terrain at speeds up to 4.25 m/s.