Bayesian Optimization for Automatic Tuning of an Energy Aware 2-D Vehicle Platoon

Vehicle platoon has the potential of improving the safety and efficiency of road transportation systems. Nevertheless, an integrated control of vehicle longitudinal and lateral motions is key to realizing these improvements. This paper focuses on developing a model predictive controller (MPC) for 2-D vehicle platoons and addresses the corresponding tuning problem of the MPC cost function hyperparameters. We adopt a distributed MPC (DMPC) framework to develop a two-stage cooperative controller, including a high-level reference generating step and a low-level stage to calculate control signals for the vehicles. The DMPC cost function trades off reference tracking versus energy consumption, therefore, the cost function tunning parameters are critical in achieving the best results. An automatic tuning framework based on Bayesian optimization (BO) is designed to search for the optimal parameters of the DMPC. Finally, the effectiveness of the proposed method is demonstrated through rigorous CarSim simulations. It is shown that the proposed methodology produces a better tracking and energy consumption compared to a baseline controller and the DMPC with predefined fixed parameters.