Mao, Jian (Northwestern Polytechnical University), ZHANG, Kailong (Northwestern Polytechnical University), Yu, Qiang (Northwestern Polytechnical University), Wu, jinfei (Northwestern Polytechnical University), Zhang, Jiahao (Northwestern Polytechnical University), de La Fortelle, Arnaud (MINES ParisTech)

A Novel Cooperative Driving Mechanism Based on Multi-Agent Deep Reinforcement Learning with Priority Parameter Sharing and Safety Rules

Scheduled for presentation during the Regular Session "Multi-autonomous Vehicle Studies, Models, Techniques and Simulations I" (ThBT10), Thursday, September 26, 2024, 14:30−14:50, Salon 18

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 14, 2024

Abstract

In the field of cooperative driving, challenges such as low safety, poor training efficiency, and obstruction of emergency vehicles remain prevalent. This study introduces a novel multi-agent reinforcement learning mechanism to tackle these issues, encompassing three pivotal algorithms: (1) The Parameter Sharing Asynchronous Advantage Actor Critic (PSA3C) Algorithm, which utilizes parameter sharing to bolster collaborative learning and optimize behavior among vehicles;(2) The Advanced PSA3C Algorithm Based on Safety Rules (PSA3C_safe) , an enhancement of PSA3C with embedded safety protocols to ensure decision-making under safety constraints; and (3) The Advanced PSA3C Algorithm based on Vehicle Priority Parameter (PSA3C_prio), which integrates priority rules to grant precedence to emergency vehicles, thereby refining traffic flow and emergency response. Our simulation tests demonstrate the efficacy of the framework, with significant improvements in critical performance indicators. Specifically, the proposed algorithms have yielded an increase in a higher average travel speed, an extension of the safe driving distance, and an elevated cumulative reward value. These enhancements underscore the potential of our approach to bolster the safety and efficiency of autonomous driving systems.