Adaptive Feed-Forward Model Predictive Control for Unmanned Vehicles

To address the challenge of simultaneously achieving accuracy at low speeds and stability at high speeds in the lateral control of unmanned vehicles, this paper proposes a feed-forward adaptive weight controller based on model predictive control (FFMPC) aimed at enhancing both accuracy and adaptability during the lateral control of these vehicles. By considering the coupling characteristics of trajectory and heading errors, we derive an error dynamics model that incorporates path curvature and its rate of change into the conventional dynamics model. The main control system is constructed using a model predictive controller with decoupling characteristics, obtained through Taylor series expansion. Additionally, a Sliding Mode Controller (SMC) based on an integral proportional-integral-derivative (PID) sliding mode surface is integrated as a feed-forward component of the control system. To prevent overfitting in the controller, we define an error tolerance threshold. The front steering compensation angle is calculated using the improved preview model, which effectively adjusts the output of the primary controller and enhances system robustness under extreme operating conditions. In this study, trajectory error minimization is employed as a key performance metric for simulation. Comprehensive simulations are conducted in the Carsim and Matlab/Simulink environment under various operating scenarios for comparison and validation. The experimental results indicate that, compared to traditional controllers, the FFMPC demonstrates superior accuracy, robustness and rapid convergence of kinematic states.