The invention provides an intersection decision-making method based on multi-agent deep reinforcement learning, and the method comprises the steps: enabling each agent to interact with an environment, obtaining observable agent information and local environment information, transmitting the overall state information to a model for learning, carrying out the centralized training of all agents, and carrying out the exploration according to the environment. And the intelligent agent gives a corresponding decision and finally transmits the decision to the bottom layer control of the intelligent agent vehicle. The current reward is obtained each time, and a new round of training is carried out according to the current reward and the next state, so that model building is completed. The method has the advantages that by improving the deep neural network model, the process of action exploration is optimized, the process of exploration of invalid actions is reduced, the robustness of a decision model is enhanced, the performance of the model in a complex scene is improved, and the training difficulty of the model is reduced. The generalization ability of the model method in a complex scene is improved through a reward function involving multiple constraints.

    本发明提供一种基于多智能体深度强化学习的交叉路口决策方法,各智能体均与环境进行交互,获取可观测到的智能体信息和局部环境信息,将整体状态信息传递给模型进行学习,对所有智能体集中训练,根据环境进行探索动作,智能体给出相应的决策,最终传输到智能体车辆的底层控制。每回得到当前奖励,根据当前奖励和下一个状态进行新回合训练,以此来完成模型搭建。优点是通过对深度神经网络模型进行改进,优化了动作探索的过程,减低对无效动作的探索过程,增强决策模型的鲁棒性,增加模型在复杂场景下的性能,降低模型的训练难度。通过涉及多约束的奖励函数,增加模型方法在复杂场景下的泛化能力。


    Access

    Download


    Export, share and cite



    Title :

    Intersection decision-making method based on multi-agent deep reinforcement learning


    Additional title:

    基于多智能体深度强化学习的交叉路口决策方法


    Contributors:
    DU YU (author) / JIANG ANNI (author) / ZHAO SHIXIN (author) / WANG YIWEI (author) / CHEN ZEYU (author)

    Publication date :

    2024-02-02


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS



    End-to-End Intersection Handling using Multi-Agent Deep Reinforcement Learning

    Capasso, Alessandro Paolo / Maramotti, Paolo / Dell'Eva, Anthony et al. | IEEE | 2021


    Vehicle-station-network interaction decision-making method based on multi-agent deep reinforcement learning

    ZHANG JIAN / XI YANJUN / WANG QIANG et al. | European Patent Office | 2023

    Free access

    Multi-intersection traffic signal control method based on deep reinforcement learning

    LIU LIJUAN / BAI GUANGMING | European Patent Office | 2023

    Free access


    Large-scale intersection management method based on multi-agent reinforcement learning

    ZHAO SHENGJIE / XUE JINWEI / DENG HAO | European Patent Office | 2023

    Free access