A feedforward neural network for the automatic control of a four-wheel-steering passenger car

An automatic control system using a feedforward neural network is proposed for four-wheel-steering passenger cars to mimic the behavior of human driver. The control system consists of two identical two-layer-feedforward networks and a feedback of the car heading deviation. One neural network acts as an emulator, and the second represents a feedforward controller. The synaptic weights of the networks are adjusted according to the generalized adaline weight adaptation algorithm. The duration of the general learning was 500 s, and no measurements were acquired. Computer simulation was carried out to evaluate the performance of the proposed system in tasks involving lane keeping on a curved roadway at different speeds, gusting side wind, and an obstacle-avoidance maneuver. The results showed that the system displayed good driving performance, and was capable of reproducing the steering performance of human driver. The system realized rapid, comfortable, and stable responses in the different tasks