A Pilot for a Robotic Vehicle System

The role of an automatic pilot in an autonomous land vehicle system is to guide the vehicle along its mission path, while at the same time avoiding unexpected obstacles. We demonstrate an algorithm that accomplishes this task in real-time with the capability to generate vehicle guidance commands every 60 milliseconds. Many aspects of obstacle avoidance and path following can be shown to depend only on the current snapshots of the local work space with only a few global parameters needed. The algorithm works by finding conical-shaped usable subsets of the local work space and is motivated by the heuristic requirement of staying in sensor view of the subgoal. Our approach allows for vehicle dynamics which is often neglected in other approaches and yields smooth paths. The problem of traps inherent in potential field approaches is avoided by introducing a winding angle principle. Before field testing, the performance of the pilot was evaluated by the computer simulation of a 10-ton tracked vehicle. During recent field tests, the pilot was successful in guiding an M113 armored personnel carrier around obstacles while staying inside the path corridor.