Scholarlink

Variable Speed Limits (VSL) have been used to dynamically control traffic via the adaptation of vehicular speed with the aim to mitigate congestion, enhance safety and traffic flow performance. Recent studies have demonstrated the effectiveness of employing Deep Reinforcement Learning (DRL) techniques for VSL controllers, resulting in improved safety, efficiency, and environmental sustainability on highways. However, the conventional RL models for VSL control systems have not been trained to prioritize and facilitate the movement of emergency vehicles. This paper proposes a deep deterministic policy gradient (DDPG) algorithm which utilizes an actor-critic model by training the DVSL controller and a novel reward function to optimize the time loss of emergency vehicles across a highway weaving section. The results show that the proposed approach outperforms the baseline method with no VSL in terms of the accumulated waiting time, accumulated time loss, average time loss of emergency vehicles and the average travel time of all the vehicles.

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

Enhancing the Performance of Emergency Vehicles Through Reinforcement Learning Based DVSL Control Systems

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Fatemeh Bandarian,
  • Eric Gyamfi,
  • Evangelos Mitsakis,
  • Evangelos Mintsis,
  • Abdollah Malekjafarian,
  • Fatemeh Golpayegani

摘要

Variable Speed Limits (VSL) have been used to dynamically control traffic via the adaptation of vehicular speed with the aim to mitigate congestion, enhance safety and traffic flow performance. Recent studies have demonstrated the effectiveness of employing Deep Reinforcement Learning (DRL) techniques for VSL controllers, resulting in improved safety, efficiency, and environmental sustainability on highways. However, the conventional RL models for VSL control systems have not been trained to prioritize and facilitate the movement of emergency vehicles. This paper proposes a deep deterministic policy gradient (DDPG) algorithm which utilizes an actor-critic model by training the DVSL controller and a novel reward function to optimize the time loss of emergency vehicles across a highway weaving section. The results show that the proposed approach outperforms the baseline method with no VSL in terms of the accumulated waiting time, accumulated time loss, average time loss of emergency vehicles and the average travel time of all the vehicles.