Arterial traffic signal control (TSC) is one of the most important issues in urban transportation system. Recent studies use Reinforcement Learning (RL) for area signal control and show superior performance, but leave several challenges to be addressed. Firstly, existing works generally construct a local RL agent for each intersection with a local objective, which may not lead to the optimum of area traffic efficiency. Secondly, most of the RL solutions adopt adaptive phase switching strategy, which is impractical and may severely affect traffic safety. Furthermore, a specific coordination pattern emerged in arterial signal control, but it is rarely studied in previous works. To address these challenges, we propose HALight, a novel RL method with hierarchical structure consisting of a manager and several workers. The manager provides sub-goals for workers according to the arterial objective, while the workers realize their sub-goals and optimize local traffic efficiency simultaneously. Moreover, HALight adopts cycle-based strategy to refine practicability and safety for real-world applications. We evaluate our method on both synthetic and real-world road networks using SUMO simulator, and experimental results show that HALight outperforms the state-of-the-art methods on both datasets.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    HALight: Hierarchical Deep Reinforcement Learning for Cooperative Arterial Traffic Signal Control with Cycle Strategy


    Contributors:
    Zeng, Jing (author) / Xin, Jie (author) / Cong, Ya (author) / Zhu, Jiancong (author) / Zhang, Yihao (author) / Jiang, Weihao (author) / Pu, Shiliang (author)


    Publication date :

    2022-10-08


    Size :

    2838290 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English






    Multi-Agent Deep Reinforcement Learning for Decentralized Cooperative Traffic Signal Control

    Zhao, Yang / Hu, Jian-Ming / Gao, Ming-Yang et al. | ASCE | 2020


    Variable lane and traffic signal cooperative control method based on deep reinforcement learning

    DING CHUAN / NIE WUYANG / LU YINGRONG et al. | European Patent Office | 2020

    Free access