Modeling a professional driver in ultra-high performance maneuvers with a hybrid cost MPC

Abstract

During racing or vehicle performance evaluation, a professional driver looks ahead and chooses the vehicle controls to best maneuver the circuit or track in minimum time. This process is carried out in a receding horizon fashion around the circuit or track in a process that closely resembles Model Predictive Control (MPC). Professional drivers quickly build mental models of the circuit or track of how best to exploit its features in order to negotiate the full maneuver in minimum time. In this article, we implement a method for exploiting future preview information into a local solution of a MPC which indirectly models a driver learning a particular circuit or track. A hybrid cost MPC structure that is capable of switching between two different driving styles is used to achieve the incorporation of this future preview information. It will be shown that this future preview information (beyond the local MPC horizon) can help to alleviate some of the sub-optimal behavior inherent in MPC. The concept is then demonstrated on a chicane maneuver. The proposed scheme is compared against the exact time-optimal control solution for this maneuver.