Applying a WNN-HMM Based Driver Model in Human Driver Simulation: Method and Test

Modeling and evaluation of human driving behavior are the core to intelligent transportations and autonomous vehicles. This paper applies a human-like driver model based on vehicle test data and the neural network. This is accomplished by compromising the merits of error-based HMM-PID module and style-based neural network algorithm, both of which will work together to form a united driver model. In the simulation, the comparisons on driving performance, e.g., fuel economy and target following ability, are presented between PID-like driver and the proposed human-like driver. Several driving behavior criteria published by SAE, e.g., energy rating and energy economy rating, are borrowed in this paper to provide standardized metrics for evaluating the driver performance on fuel economy and emissions. Experimental results verified the effectiveness of the proposed scheme.