Global optimization of local weights in mixed-cost mpc for minimum time vehicle maneuvering

Abstract

In this paper a Model Predictive Control (MPC) strategy is utilized to model a professional driver negotiating a set driving circuit in minimum time. MPC is inherently suboptimal because not all future information is incorporated into each horizon of the control scheme. Motivated by how professional drivers learn race circuits in order to best exploit its features, we will alleviate some of the suboptimality inherent to MPC by optimizing the local cost function of each MPC horizon. This will allows objectives over a local segment to be properly adjusted such that the global goal of minimizing maneuvering time over a full maneuver is approximated. This problem is solved utilizing a cascaded optimization structure with the inner loop recursively solving the MPC problem around the track and the outer loop optimizing the weights of the local MPC horizons. It will be shown that by varying weights at key locations on a particular maneuver, performance gains can be realized compared to a traditional time optimal MPC strategy.