Hu, Zhaofeng (Stony Brook University), Wang, Xu (Beijing jiaotong University), Wang, Zongyao (Dalian Maritime University)

Enhancing Urban Intersections Management through Deep Reinforcement Learning: Superior Control of Autonomous Vehicles in Mixed Traffic Flows

Scheduled for presentation during the Regular Session "Autonomous vehicles - intersection management" (FrAT8), Friday, September 27, 2024, 12:10−12:30, Salon 16

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 8, 2024

Abstract

Urban intersections frequently become hotspots for traffic congestion. Traditional traffic signal control methods are rigid and often ineffective at alleviating these congestion issues. In recent years, the emergence of mixed traffic flows, including both Conventional human-driven Vehicles (CVs) and autonomous vehicles (AVs), has provided new possibilities for addressing congestion at intersections. This paper presents a control algorithm for autonomous vehicles based on deep reinforcement learning within mixed traffic scenarios, which abandons traditional traffic signals and demonstrates superior control performance. Through extensive experimentation, the effectiveness of the algorithm has been validated: the experimental results show that at a 50% penetration rate of autonomous vehicles, our algorithm significantly reduces average waiting times and increases flow speeds compared to traditional traffic signal methods, indicating improved stability and efficiency in traffic management. Additionally, stress tests in extreme traffic scenarios were conducted, and the results reveal that even at a 50% AV penetration rate, our algorithm still outperforms traditional traffic signal control methods.