The performance of the model-based controller is always affected by the uncertainty and nonlinearity of the model parameters in the vehicle path tracking process. To address this issue, a novel path tracking controller based on model-free adaptive dynamic programming (ADP) is proposed for autonomous vehicles in this paper. To be specific, the proposed controller obtains information from the online state and front-wheel angle input data which are repeatedly used to calculate the controller gain iteratively. So, this controller features not requiring accurate knowledge of vehicle model parameters for controller development. Meanwhile, the path tracking performance of the autonomous vehicle will be inevitably disturbed by unknown nonlinear external disturbance. To approximate this disturbance, the learning characteristics of Radial Basis Function Neural Network (RBFNN) are applied to generate compensation for the front-wheel angle. Afterward, the weight updating law of RBFNN is derived by Lyapunov function to ensure the stability and convergence of the whole system. Finally, Hardware in the loop (HIL) test results demonstrate that the proposed ADP-RBF controller can improve the comprehensive performance of the vehicle path tracking control system and achieve the balance between path tracking accuracy and minimum sideslip angle.

在车辆路径跟踪过程中，基于模型的控制器的性能总是受到模型参数的不确定性和非线性的影响。为了解决这个问题，本文为自动驾驶汽车提出了一种基于无模型自适应动态规划（ADP）的新型路径跟踪控制器。具体来说，所提出的控制器从在线状态和前轮角度输入数据中获取信息，这些数据被重复用于迭代计算控制器增益。因此，该控制器的特点是不需要准确了解用于控制器开发的车辆模型参数。同时，自主车辆的路径跟踪性能将不可避免地受到未知非线性外部干扰的干扰。为了近似这种扰动，应用径向基函数神经网络 (RBFNN) 的学习特性来生成前轮角度补偿。之后，通过李雅普诺夫函数推导出RBFNN的权重更新规律，以保证整个系统的稳定性和收敛性。最后，硬件在环 (HIL) 测试结果表明，所提出的 ADP-RBF 控制器可以提高车辆路径跟踪控制系统的综合性能，并实现路径跟踪精度和最小侧滑角之间的平衡。