基于PID与TD3残差混合策略的自动驾驶路径跟踪控制研究OA

In order to solve the problems of poor parameter adaptability of traditional feedback controller under nonlinear conditions such as continuous large curvature and low safety caused by blind exploration of end-to-end deep reinforcement learning(DRL)in the early stage of training,a residual hybrid control method based on PID and twin delayed deep deterministic policy gradient(TD3)is proposed.The stability of the system is guaranteed by the basic output of the PID controller,and the feedforward nonlinear residual compensation is carried out by using the strategy learning ability of the TD3 agent.The augmented state space of the fusion path curvature preview and PID internal state is designed.A multi-objective reward function considering path track-ing accuracy,comfort and safety is constructed.The simulation test in CARLA shows that this method success-fully solves the problem of phase lag of pure PID control,and reduces the root mean square error of lateral tracking by 87.1%.Compared to the pure end-to-end TD3 algorithm,the steering wheel command change rate decreases by more than 90.1%,effectively suppressing high-frequency jitter phenomenon.The results show that the method realizes the decoupling of physical safety and tracking optimality,and provides a reliable solution for the path tracking control of autonomous driving.