The invention discloses an automatic driving system backdoor attack method based on deep reinforcement learning and a related device. The method comprises the following steps: determining a threat model according to attacker ability and a target; under the threat model, determining a state space, an action space and a reward function of the deep reinforcement learning model, and designing a malicious reward function of backdoor attack; designing a backdoor trigger on the basis of the malicious reward function, and hiding the backdoor trigger in a series of continuous space-time states; and fusing the malicious reward function and the backdoor trigger into the training process of the deep reinforcement learning model, configuring training parameters, and training and deploying the deep reinforcement learning model with the backdoor. According to the method, the backdoor attack is performed on the deep reinforcement learning by using the time and space characteristics of vehicle driving, the attack success rate is higher, and the performance of the automatic driving system can be influenced to a lower extent under the condition that a trigger does not exist.

    基于深度强化学习的自动驾驶系统后门攻击方法及相关装置,包括:根据攻击者能力和目标确定威胁模型;在威胁模型下,确定深度强化学习模型的状态空间、动作空间和奖励函数,同时设计后门攻击的恶意奖励函数;在恶意奖励函数的基础上,设计后门触发器,将其隐蔽到一系列连续的时空状态中;将恶意奖励函数以及后门触发器融入到深度强化学习模型的训练过程中,配置训练参数,训练并部署带有后门的深度强化学习模型。本发明利用车辆驾驶的时间和空间特征对深度强化学习进行后门攻击,具有更高的攻击成功率,且在触发器不存在的情况下,能够更低限度地影响自动驾驶系统的表现。


    Access

    Download


    Export, share and cite



    Title :

    Automatic driving system backdoor attack method based on deep reinforcement learning and related device


    Additional title:

    基于深度强化学习的自动驾驶系统后门攻击方法及相关装置


    Contributors:
    LIU JIAJIA (author) / YU YINBO (author) / YAN SAIHAO (author)

    Publication date :

    2023-07-04


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    H04L TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION , Übertragung digitaler Information, z.B. Telegrafieverkehr / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS



    Automatic driving decision-making method based on deep reinforcement learning

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

    Free access

    Automatic driving decision planning method based on deep reinforcement learning and deep learning

    YANG LU / ZHANG HAO / TAN YANSONG et al. | European Patent Office | 2024

    Free access

    Automatic driving decision-making system based on deep reinforcement learning

    ZHENG XIAOYAO / YAO QINGHE / ZHANG JIANPENG et al. | European Patent Office | 2024

    Free access

    Automatic driving behavior decision-making method based on deep reinforcement learning

    YANG MINGZHU / LIU XIANGWEI / LI ZHUOLUO | European Patent Office | 2020

    Free access

    Automatic driving behavior decision-making method based on deep reinforcement learning

    ZOU JUNYI / WANG HAIXIN | European Patent Office | 2024

    Free access