<p>High-precision path tracking for autonomous vehicles remains challenging because vehicle dynamics are strongly nonlinear and time-varying, which can undermine prediction accuracy and closed-loop performance in complex driving scenarios. To address these challenges, this paper presents a data-driven adaptive model predictive control (MPC) framework for autonomous vehicles path tracking. First, a real-time vehicle-parameter estimator based on the unscented Kalman filter (UKF) is developed to calibrate sensor errors and accurately infer unmeasurable states and parameters. Then, we design a data-driven adaptive MPC scheme that integrates deep deterministic policy gradient (DDPG) reinforcement learning with a multilayer perceptron (MLP) network to capture complex nonlinear dynamics. The controller further adopts an online-offline dual-optimization strategy that combines a linear residual structure with a linear time-varying (LTV) model, enabling accurate state prediction while preserving real-time computational efficiency. In addition, a vehicle-speed-adaptive prediction-horizon mechanism is incorporated to improve adaptability across varying driving conditions. Simulation results show that the proposed approach reduces tracking error, produces smoother control actions, and improves robustness relative to conventional and adaptive MPC methods, demonstrating its effectiveness for high-precision autonomous vehicle path tracking.</p>