Multi-agent Collaborative Adaptive Cruise Control Based On Reinforcement Learning

With the development and application of intelligent networked vehicles, Adaptive Cruise Control (ACC) is gradually developed and upgraded to Cooperative Adaptive Cruise Control (CACC). However, in the CACC system, following error, relative speed and acceleration have a certain range of constraints. If the acceleration fluctuates and amplifies backwards along the queue and violates the saturation constraint boundary, the acceleration of the vehicle behind will increase continuously, and the system will lose the stability of the queue and cause traffic accidents. Therefore, based on the Model Predictive Control (MPC) theory, this paper takes the CACC system as the research object and adopts the form of weighted quadratic performance functional and linear matrix inequality constraints. The design problem of the ACC is finally transformed into an online convex quadratic programming problem with constraints. At the same time, the saturation control of the CACC system is studied based on the parametric Lyapunov equation to improve the stability of the system.