Abstract
Using the results of lane-change tests performed on a driving simulator, driver models were developed by means of a neural-network system. Several kinds of driver visual information were used as input data for structure of the neural network, and the steering angle was employed as learning information. A series of simulations employing the trained neural network was conducted to determine the allowable ranges of initial vehicle position, velocity, stability factor, and other variables for successful lane-change maneuvers and for stable running against side winds.