A nonlinear approach to vehicle guidance

We present a nonlinear dynamic approach to lateral and longitudinal intelligent vehicle guidance on German “autobahnen”. The methodology resembles the dynamic neural field architecture where information is coded in the location of activation within dynamic fields. The time course of the control parameters is obtained by tracking the attractors in the output field. We assume the sensors are able to measure the distance and the time-to-collision of objects in front and behind of the controlled car on both lanes, and so radar and/or vision based sensors can be applied. This setting implies a high dimensional input space. The system is designed to make reliable decisions depending on the detected objects and the nominal velocity in the presence of disturbed sensory information. The control process results in typical driving behavior like lane keeping, overtaking, and collision avoidance