Jiang, Yongzhi (Beihang University), zhou, bin (BUAA), He, Shan (Beihang University), Li, Yongwei (Beihang University), Wu, Xinkai (Beihang University)

Decision-Making Based on Multi-Agent Reinforcement Learning for Autonomous Vehicles in Narrow Lane Meeting Scenario

Scheduled for presentation during the Regular Session "Multi-autonomous Vehicle Studies, Models, Techniques and Simulations II" (FrAT10), Friday, September 27, 2024, 11:10−11:30, 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 3, 2024

Abstract

Decision-making in narrow lane meeting scenarios remains a challenge for autonomous vehicles. It is difficult for autonomous vehicles to identify the intentions of oncoming vehicles without vehicle-to-vehicle communication. In this paper, we propose a collision avoidance decision-making method based on multi-agent reinforcement learning (MARL), which employs a multi-agent proximal policy optimization (MAPPO) method to deal with the two-vehicle game problem in narrow lane meeting scenarios. The proposed method evaluates the rewards for each vehicle to execute a specific action and the possible intention of the opponent vehicle. It allows autonomous vehicles to consider both the costs of their own actions and intentions to interact with others when making decisions. We simulate scenarios with different obstacle conditions aimed at training autonomous vehicles to make decisions to travel safely and efficiently. Experimental results show that our method can enable the autonomous vehicle and the opposing vehicle to efficiently pass through the narrow lane, with a success rate of over 97%.