Scene recognition and navigation capabilities for lane changes and turns in vision-based vehicle guidance

The development of autonomous visual guidance for land vehicles at UniBwM is reviewed. An initial breakthrough in performance level was achieved in 1986 by combined differential/integral representations of the road skeleton and the vehicle state in conjunction with recursive estimation techniques (4-D approach). Feedback control laws allowed easy implementation of reactive behavior; by complementing this approach with generic feedforward control-time histories for several mission elements, a capability for mission performance is introduced. Lane changes, turn-offs and handling forks in the road are the most essential such elements; in conjunction with a capability for landmark recognition and digital map reading this allows the autonomous performance of entire missions. Recently developed capabilities of the VaMoRs test vehicle are discussed.