Study of driver factors affecting the directional response of an articulated vehicle using neural networks

A two-layer neural network with 12 neurons in the first layer is applied to model the directional control behaviour of a driver-articulated vehicle sy stem. Six different training schemes are employed using combinations of inp ut variables related to visual perception of tracking error in terms of pos ition and orientation, and motion perception of the driver. The network is trained using the results derived from a comprehensive closed-loop analytic al model under an obstacle avoidance manoeuvre. The input data to the neura l network included: the coordinates of the previewed path, lateral position , velocity, and acceleration of the tractor and lateral acceleration of the trailer. The front wheel steer angle was considered as the desired output of the trained network. The effectiveness of the trained network describing the driver behaviour is investigated under different directional maneuvers of varying severity. It is concluded that the neural network can be traine d to model the directional control behavior of the driver, while the effect iveness of the trained network depends upon the input variables supplied to train the network. The results show that the visual perception of the driv er based on the position error contributes most significantly to the traini ng of the network. The motion perception variables in terms of lateral acce lerations of the articulated vehicle yield a more effective neural network model of the driver.