Estimation of longitudinal speed profile of car drivers via bio-inspired mirroring mechanism

This paper deals with the problem of the prediction of driver intention. The problem is relevant in the context of modern Advanced Driver Assistance Systems. More specifically, we address the task to continuously generate a predicted longitudinal velocity profile with a fixed time horizon and associated with a driver's intention (e.g. overtake). The objective is to obtain a “general purpose” prediction, aimed to feed any ADAS algorithm requiring future longitudinal velocity and intention informations, like safety applications, warning systems or MPC-based algorithms. The prediction makes use of the artificial co-driver concept, which is here designed to deal with longitudinal inputs only. The co-driver is an agent able to perform inference of intention by means of a mirroring approach, trying to imitate the human driving behavior. The approach is conceived to be simple and modular, using only longitudinal informations from the vehicle, and flexible to the availability of external informations (e.g. vehicle ahead). The works includes the implementation of a jerk filtering technique proposed by some of the authors, this technique is used in a mirroring approach for the first time. Preliminary results on prediction are presented, and future development and validation are discussed.