为了降低延迟对纯跟踪控制器的影响，提高纯跟踪控制器控制自动驾驶车辆沿规划轨迹行驶的准确性，提出了一种基于深度强化学习的纯跟踪控制器优化方法，利用深度确定性策略梯度(DDPG)模型实时预测车辆位置误差信息，动态控制纯跟踪控制器求得的转向控制信号与规划轨迹航向角信号的融合比例，以获得最佳的转向角控制信号。随机路径条件下的MATLAB仿真结果表明，使用DDPG模型调整转向控制信号与规划轨迹航向角信号的融合比例能有效提升纯跟踪控制器的控制效果，改进后的纯跟踪控制器控制车辆以介于1 m/s ~ 5 m/s之间的速度沿规划轨迹行驶时，车辆位置误差不超过0.2 m，航向角误差不超过0.1 rad，相对传统纯跟踪控制器，优化后的纯跟踪控制器控制车辆沿规划轨迹移动的横向误差减少了80%，航向角误差降低了90%。

To mitigate the impact of latency on the Pure Pursuit controller and enhance its accuracy in guiding autonomous vehicles along planned trajectories, this study proposes an optimization method for the Pure Pursuit controller based on deep reinforcement learning (DRL). Specifically, a Deep Deterministic Policy Gradient (DDPG) model is employed to predict real-time vehicle position errors and dynamically adjust the fusion ratio between the steering control signal derived from the Pure Pursuit controller and the planned trajectory's heading angle signal. This approach aims to optimize the steering angle control signal. Simulation experiments conducted in MATLAB under random path conditions demonstrate that the DDPG-based adaptive fusion mechanism significantly improves the control performance of the Pure Pursuit controller. For a vehicle traveling at speeds ranging from 1 m/s to 5 m/s along the planned trajectory, the optimized controller achieves a maximum position error of 0.2 m and a heading angle error of 0.1 rad. Compared to the traditional Pure Pursuit controller, the proposed method reduces lateral errors by 80% and heading angle errors by 90%, thereby validating its effectiveness in enhancing trajectory-tracking precision.