Model-based neural distance control for autonomous road vehicles

In this paper, a model-based neural distance controller is presented which directly gives control signals to throttle and brake. The neural network itself consists of a simple multilayer feed forward perceptron network. A special training method is used where the neural network is trained on a detailed nonlinear dynamic longitudinal vehicle model by minimizing a cost function. Only a few simulated driving manoeuvres are necessary to train the controller. Practical road tests with the Daimler-Benz experimental vehicle OSCAR (MB 300 TE station wagon) show that the model-based neural distance controller can be used for intelligent autonomous cruise control as well as for distance control in stop & go-traffic.