Autonomous Racing Car Controller Based on Gaussian Process Residual Error Prediction Model

This study aims to solve the control problem of autonomous vehicles under specific conditions at high speeds, which has significant implications for enhancing the safety and performance of autonomous vehicles in real-world high-speed scenarios, such as in motorsports or emergency maneuvers in urban environments. We constructed an accurate model of the vehicle to cope with parameter mismatch and uncertainty perturbation using Gaussian process for the typical conical bucket environment for a racing car. An ideal three-degree-of-freedom vehicle dynamics model is first established, and the Gaussian process is utilized to learn and predict the residual error for model parameter mismatch and uncertainty perturbation. Then the Gaussian process residual error model is applied to the nonlinear model predictive controller (MPC) to form the Gaussian process nonlinear model predictive controller (GPNMPC), and finally the simulation results show that the robustness under uncertain disturbances is improved.