Generation of Pedal Operation Patterns of Individual Drivers in Car-Following for Personalized Cruise Control

This paper presents a method to generate car-following patterns for individual drivers. We assume that driving is a recursive process. A driver recognizes a road environment such as velocity and following distance and adjusts gas and brake pedal positions. A vehicle status changes according to the driver's operation and the road environment changes according to the vehicle status. Driving patterns of each driver are modeled with a Gaussian mixture model (GMM), which is trained as a joint probability distribution of following distance, velocity, pedal position signals and their dynamics. Gas and brake pedal operation patterns are generated from the GMMs in a maximum likelihood criterion so that the conditional probability is maximized for a given environment i.e., following distance and velocity. Experimental results for a driving simulator show that car-following patterns generated from GMMs for three different drivers maintain their individual driving characteristics.