Steering control for car cornering by means of learning using neural network and genetic algorithm

Car drivers learn steering operation with exercises, but car dynamics is nonlinear at high speed situation on rough roads or low friction roads. Although skillful drivers might control cars for such nonlinear dynamics, it is difficult for ordinary drivers to control cars for such a situation. In this paper, steering operation for cornering is learned by using a neural network (NN) and a genetic algorithm (GA). The NN controller drives car autonomously with visual information and car states. The inputs to the NN controller are the direction and the curvature of the object path, and the lateral position, the yaw rate and the slip angle of the car. The output from the NN controller is the front steering angle. 4 wheel nonlinear car model with the magic formula of pure cornering is used for an analytical model. The NN controller acquires the driving operation on the curved road as a result of 30 generations iteration of the GA learning. It drives the car successfully on learned and non-learned curved roads. And, it shows the operation that is similar to the counter steering operation which is used by World Rally Championship drivers at tight curved roads. It achieves higher manoeuvrability than any other positive steering controller by using the counter steering. As a result, the availability of the NN controller learns by the GA algorithm for vehicle autonomous driving in nonlinear region is shown. Copyright (C) 1998 IFAC.