Model predictive control combined reinforcement learning for automatic vehicles applied in intelligent transportation system

Abstract

This paper presents the design of model predictive control (MPC) combined reinforcement learning (RL) applied in an intelligent transportation system (ITS). The car is to follow the reference path by the MPC control, and its parks in the parking by RL has been trained. The MPC controller constantly moves the vehicle along the reference path while the MPC algorithm searches for an empty parking spot. Meanwhile, the reinforcement learning-proximal policy optimization (RL-PPO) control will perform parking on demand if the MPC finds a parking position. This hybrid controller can quickly implement programming on MATLAB software by writing code. Furthermore, this hybrid controller simultaneously performs precise detection and avoidance of obstacles in tight parking spaces. The correctness of the theory is demonstrated through MATLAB/Simulink.