MPC-inspired neural network method for cooperative control of vehicle platoons

Abstract

A model predictive control (MPC) inspired neural network (NN) method is proposed to solve the cooperative problems of vehicle platoon in this paper. The controller design is approximate to a quadratic programming (QP) solver for MPC problems. However, the method proposed in this paper is based on data-driven MPC rather than being strictly model-based. Meanwhile, compared to QP solver, the computational efficiency is significantly improved. To ensure asymptotic convergence of the vehicle platoons, terminal penalty matrix and terminal set are taken into the optimization problem, and a supervised learning based feedforward neural network is trained to approximate control inputs. Compared to traditional neural network controllers, this method has the similar performance of ensuring asymptotic stability as model predictive controller. To validate the effectiveness of the proposed controller, the information from the leading vehicle utilizes real-world driving data from commercial trucks, which more accurately reflects the dynamic behavior of the vehicle under actual driving conditions. Based on the real truck platform, the experimental results show that when the leading vehicle accelerates or decelerates, the following vehicles in the platoon can make real-time responses while exhibiting excellent dynamic performance. In addition, joint simulations based on MATLAB and Trucksim in the curved road scenario show that the performance of lane keeping can also be guaranteed.