Abstract
This paper describes an analysis of control algorithm for drivers. At first step for this research, the driver model is constructed using the neural network system. The data on two courses are used for learning. One has a record for a lot of accidents, and the other has few. It is found that the model has good description of the driver's control manner adequately. Next, the constructed driver model is evaluated using two analytical methods. And it is shown that the control algorithm and environmental recognition are much different between two courses.