Approximated Long Horizon MPC with Hindsight for Autonomous Vehicles Path Tracking

Abstract

We propose an approximated long horizon model predictive control (MPC) for path tracking of autonomous vehicles, which is more computationally efficient than a standard MPC with a long horizon and more effective than a standard MPC with a short horizon. In the proposed MPC, the cost function consists of two parts: 1) the cost function of the short horizon MPC, and 2) an additional term to approximate the difference between the cost function with the short horizon and that with the long horizon, which we call the hindsight cost function. The additional term is obtained from a linear regression model that is offline learned from previous known trajectory data. Finally, a CarSim-MATLAB/Simulink co-simulation is provided to show the effectiveness of the proposed approximated long horizon MPC.