Zhao, Jinghua (JiLin Normal University)

Research on Car-Following Strategy on Dynamic Scene Based on Reinforcement Learning Technology

Scheduled for presentation during the Invited session "Learning-based Optimization and Control for Intelligent Electrified Vehicles" (ThB1), Thursday, October 31, 2024, 14:10−14:30, Room T1

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on May 16, 2025

Abstract

A novel car-following control method based on reinforcement learning technology is proposed in this paper. The Actor-Critic framework of reinforcement learning is adopted by this method, which offline calibrates the initial policy neural network parameters based on experimental data, and online updates the policy neural network parameters based on Bellman theory. The experimental results of multiple transient operating conditions in the MATLAB/CARSIM joint simulation environment show that the method proposed can achieve the approximate car-following performance of the PID controller, and can simultaneously meet multiple goals such as energy conservation, car-following and comfort. The ability of parameter adaptive updating ensures the generalization performance of the control method proposed.