Efficient Self-driving Control for Electric Vehicles

Technology advancement of smart mobility not only contributes to energy saving but also driving efficiency. Particularly, self-driving methodology with echo-driving mode has become increasingly crucial due to the inherent limitations of power capacity in electric vehicles (EVs). This research paper introduces a hybrid control algorithm that combines Model Predictive Control (MPC) and Deep Reinforcement Learning (DRL) to improve the energy efficiency of battery powered EVs. In this paper, the MPC controller is devised to generate speed profiles for the EVs, aiming to maximize energy efficiency within a given receding horizon. The terminal cost associated with this horizon is then incorporated into the DRL module as observation information at a state. Consequently, the optimal cost or optimal control strategy for a given state is learned from its nearest states, and the previous state is constraints for the next state and terminal condition. The MPC-DRL algorithm was verified using AirSim which is open-source autonomous vehicle simulator, and the simulation results demonstrate its effectiveness in improving energy efficiency.