The invention belongs to the field of deep reinforcement learning automatic driving, and relates to a vehicle safety decision-making technology, in particular to a cascade deep reinforcement learning safety decision-making method based on multi-modal space-time characterization, which comprises the following steps of: firstly, constructing a multi-modal space-time perception encoder to combine modeling space and motion information from multi-modal continuous input; obtaining a current perceptual representation of the dynamic driving scene; then, a future prediction encoder is introduced to capture interaction among different traffic participants from the current perception representation, and future prediction representation is obtained; and then, the current perception representation and the future prediction representation are connected to form a multi-modal space-time representation which is used as state input of reinforcement learning so as to comprehensively grasp a scene, and in combination with a distributed PPO algorithm, a security decision task is realized under the guidance of a reward function for security decision design. The method has very high environmental adaptability and decision success rate, and can realize an active safety decision task of the intelligent automobile in a dense traffic scene and under emergencies.

    本发明属于深度强化学习自动驾驶领域,涉及车辆安全决策技术,具体是一种基于多模态时空表征的级联深度强化学习安全决策方法,其先构建了一个多模态时空感知编码器从多模态连续输入中联合建模空间和运动信息,以获取动态驾驶场景的当前感知表征;而后,引入未来预测编码器从当前感知表征中捕获不同交通参与者之间的交互,获取未来预测表征;而后,连接当前感知表征和未来预测表征形成多模态时空表征并作为强化学习的状态输入,以全面把握场景,并结合分布式PPO算法,在针对安全决策设计的奖励函数指导下实现安全决策任务。本发明具有很高的环境适应性和决策成功率,能够在稠密交通场景中以及突发事件下实现智能汽车的主动安全决策任务。


    Access

    Download


    Export, share and cite



    Title :

    Cascade deep reinforcement learning security decision-making method based on multi-modal space-time representation


    Additional title:

    基于多模态时空表征的级联深度强化学习安全决策方法


    Contributors:
    YANG YUXIANG (author) / GE FENGLONG (author) / ZHAO JUFENG (author) / FAN JINLONG (author) / DONG ZHEKANG (author) / GAO MINGYU (author)

    Publication date :

    2024-10-29


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung / B60W CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION , Gemeinsame Steuerung oder Regelung von Fahrzeug-Unteraggregaten verschiedenen Typs oder verschiedener Funktion / G01D MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE , Anzeigen oder Aufzeichnen in Verbindung mit Messen allgemein / G01S RADIO DIRECTION-FINDING , Funkpeilung / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G06V



    Deep reinforcement learning-based ethical-driven multi-modal decision-making method

    LI XUEYUAN / GAO XIN / LUAN TIAN et al. | European Patent Office | 2023

    Free access

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

    DU YU / JIANG ANNI / ZHAO SHIXIN et al. | European Patent Office | 2024

    Free access


    Lane keeping decision-making method based on deep reinforcement learning

    WANG PENGWEI / ZHOU HENGHENG / GAO SONG et al. | European Patent Office | 2023

    Free access

    Automatic driving decision-making method based on deep reinforcement learning

    LIU CHENGQI / LIU SHAOWEIHUA / ZHANG YUJIE et al. | European Patent Office | 2024

    Free access