【作者】 潘之瑶； 郑宏宇； 郭中阳； 陈晨；

【Author】 Pan Zhiyao;Zheng Hongyu;Guo Zhongyang;Chen Chen;Jilin university;Jiangsu Chaoli Electric Manufacture Co.,Ltd.;

【机构】 吉林大学； 江苏超力电器有限公司；

【摘要】 本文提出了一种基于深度强化学习(DRL)的车辆动态避障方法,用于处理存在多个动态障碍物的复杂避障场景。首先,建立了车辆运动学模型、车道线模型、任务奖励函数以及车辆状态空间和动作空间的数学模型。其次,采用Singer模型作为障碍物运动状态预测的基础,并结合扩展卡尔曼滤波(EKF)算法形成Singer-EKF方法。之后,本文利用深度确定性策略梯度(DDPG)算法在解决连续动作空间下序列决策问题的优势,设计了车辆动态避障方法。最后,通过仿真实验对本文所提出方法的可行性和有效性进行了验证。结果表明,Singer-EKF方法通过有效结合实际测量数据和Singer模型预测数据,能够精准预测动态障碍物的运动轨迹,为车辆在巡航过程中的动态避障策略提供了关键的输入;基于DDPG算法的动态避障方法能够使被控车辆在复杂的避障环境下成功完成对多个动态障碍物的规避并保持稳定状态。更多还原

【Abstract】 This paper proposes a vehicle dynamic obstacle avoidance method based on deep reinforcement learning (DRL) to handle complex obstacle avoidance scenarios with multiple dynamic obstacles.Firstly,mathematical models are established for the vehicle′s kinematics,lane model,task reward function,as well as the vehicle′s state space and action space.Secondly,the Singer model is adopted as the foundation for predicting the motion state of obstacles,combined with the extended Kalman filter (EKF) algorithm to form the Singer-EKF method.Next,leveraging the advantages of the deep deterministic policy gradient (DDPG) algorithm in solving sequential decision-making problems in continuous action spaces,the vehicle dynamic obstacle avoidance method is designed.Finally,simulation experiments are conducted to validate the feasibility and effectiveness of the proposed method.The results demonstrate that the Singer-EKF method accurately predicts the motion trajectories of dynamic obstacles by effectively combining actual measurement data with Singer model prediction data,providing crucial input for the vehicle′s dynamic obstacle avoidance strategy during cruising.The dynamic obstacle avoidance method based on the DDPG algorithm enables the controlled vehicle to successfully evade multiple dynamic obstacles in complex avoidance environments while maintaining stability.更多还原