Nonlinear model predictive control for lateral path tracking of intelligent vehicles considering model uncertainties

Under critical maneuvers like high-speed emergency steering, conventional Linear Model Predictive Control (LMPC) suffers from insufficient path-tracking accuracy, owing to coupled effects of multiple sources of model uncertainties -including model simplification errors, parameter perturbations, and unmodeled dynamics, as well as a limited capability to characterize highly nonlinear vehicle dynamics, which may even lead to safety issues like trajectory deviation or instability of intelligent vehicles. To tackle these challenges, this paper proposes a Model Uncertainty-Nonlinear Model Predictive Control (MU-NMPC) algorithm for lateral path tracking that accounts for model uncertainties. The framework initiates by establishing a Nonlinear Model Predictive Control (NMPC) structure using a nonlinear two-degree-of-freedom (2DOF) vehicle model, which is solved via Sequential Quadratic Programming (SQP). Furthermore, a state-error-based uncertainty compensation mechanism is developed to mitigate model uncertainties through online closed-loop feedback identification and real-time filtering for prediction errors estimation. Finally, the efficacy of the proposed MU-NMPC algorithm is verified through both simulations tests and SpeedGoat-based rapid control prototyping (RCP) experiments. Compared with LMPC, MU-NMPC exhibits superior performance under extreme conditions: it not only maintains sufficient control margins and enhances the vehicle’s emergency handling capability but also significantly improves path-tracking accuracy. Moreover, in the presence of parameter variations such as ±20% perturbations in vehicle mass and moment of inertia, MU-NMPC effectively suppresses the rate of change in path tracking error, demonstrating strong robustness performance in complex operational scenarios.