An Adaptive MPC Based on RL for Vehicle Longitudinal Control

Model Predictive Control (MPC) is extensively utilized in contemporary advanced control systems due to its ability to perform optimization and handle constraints effectively. However, the performance of MPC heavily relies on model pre-cision and the operating environment, making it less effective in rapidly changing dynamic conditions. To address this limitation, a hybrid algorithm combining Reinforcement Learning (RL) with MPC is proposed. This approach leverages RL to adaptively tune the parameters of MPC based on environmental changes, enhancing its ability to handle dynamic scenarios. To validate the effectiveness of this hybrid algorithm, it was applied to a vehicle braking scenario. Experimental results demonstrate that the combined RL-MPC algorithm significantly improves the performance of traditional MPC, especially in environments with changing conditions.