Abstract: A self-adaptive smooth model predictive control (MPC) method was proposed to address the issues of large tracking errors and steering abruptness in vehicle lateral control. Firstly, an augmented vehicle dynamics model incorporating errors in lateral position, heading angle, sideslip angle, and yaw rate was established, and an adaptive mechanism was introduced to dynamically adjust the sampling period, prediction horizon, and control horizon. Secondly, linearly increasing smooth weighting terms for the control inputs and their rates of change were incorporated into the objective function to suppress steering angle fluctuations and improve control smoothness, after which the quadratic programming method was employed to solve for the optimal front-wheel angle control sequence in real time. Through co-simulation using CarSim and Simulink, the proposed method was compared with traditional MPC and LQR controllers under three vehicle speeds of 36, 54, and 72 km/h and various road adhesion conditions. The results demonstrate that the improved MPC controller significantly outperforms both the traditional MPC and LQR controllers across all tested conditions. The maximum lateral tracking error is reduced by 37% to 70%, while the generated front-wheel steering angle commands are smoother and more continuous, effectively avoiding high-frequency chattering and actuator overload. Stable path tracking capability is maintained even on low-adhesion road surfaces (μ=0.3), which validates the excellent control accuracy and smoothness of the proposed method under various operating conditions. By synergistically optimizing the prediction model and the objective function, a balance between high-precision trajectory tracking and control smoothness is effectively achieved, providing a highly robust control solution for autonomous vehicles under varying working conditions.