Evaluation of driver-behavior models in real-world car-following task

In this paper, we investigate the performance of recently proposed driver-behavior modeling techniques for car-following task based on Gaussian mixture model (GMM) and piecewise auto regressive exogenous (PWARX) algorithms. Both driver-behavior modelings are employed to anticipate car-following driving behavior in terms of pedal control behavior (brake and gas/accelerator pedal operation) in response to the observable driving signals of vehicle behavior (i.e., vehicle velocity and following distance between vehicles). The evaluation is conducted using real-world driving data obtained from several drivers under a variety of driving environments. We illustrate the prediction capability of both representative models as the anticipatory time increases from one to five seconds. Furthermore, we investigate the driver-behavior model adaptation framework as a means to better extract individual driving characteristics when the amount of individual driving data is inadequate. This evaluation provides a valuable understanding driver-behavior models' characteristics of real-world car following and its potential to prevent rear-end collision.