Scholarlink

As a typical traffic bottleneck, urban intersections are usually high-incidence areas for traffic congestion and traffic accidents. Instead of the classic approach to developing an adaptive traffic light, a novel approach is introduced to the urban intersection control problem where the speed limits of intersection approaches are regulated cooperatively in a dynamic fashion. The primary objective of this paper is the smoothing the turbulent traffic flow to reduce traffic conflicts in the urban intersection using Variable Speed Limit control (VSL) control. A Multi-Agent Reinforcement Learning (MARL) framework is utilized to get the optimal speed limits. The control effect was tested by applying high-fidelity microscopic traffic simulation (SUMO). Moreover, the traffic conflicts were evaluated through the Surrogate Safety Assessment Model (SSAM). Simulation results show that the proposed MARL-based VSL control framework is a potential, effective control method to smooth traffic dynamics in general and to reduce accident risk at urban intersections. Compared with existing traffic safety strategies, our control method reduces the total traffic conflicts by 9.39%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-agent Reinforcement Learning-Based Traffic Control for Urban Intersections Using Variable Speed Limit Control

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Xuan Fang,
  • Tamás Tettamanti

摘要

As a typical traffic bottleneck, urban intersections are usually high-incidence areas for traffic congestion and traffic accidents. Instead of the classic approach to developing an adaptive traffic light, a novel approach is introduced to the urban intersection control problem where the speed limits of intersection approaches are regulated cooperatively in a dynamic fashion. The primary objective of this paper is the smoothing the turbulent traffic flow to reduce traffic conflicts in the urban intersection using Variable Speed Limit control (VSL) control. A Multi-Agent Reinforcement Learning (MARL) framework is utilized to get the optimal speed limits. The control effect was tested by applying high-fidelity microscopic traffic simulation (SUMO). Moreover, the traffic conflicts were evaluated through the Surrogate Safety Assessment Model (SSAM). Simulation results show that the proposed MARL-based VSL control framework is a potential, effective control method to smooth traffic dynamics in general and to reduce accident risk at urban intersections. Compared with existing traffic safety strategies, our control method reduces the total traffic conflicts by 9.39%.