Model Predictive Controller Design Based on Residual Model Trained by Gaussian Process for Robots

Model mismatch is inevitable in robot control due to the presence of unknown dynamics and perturbations. Traditional model predictive algorithms are usually based on constant value assumptions not able overcome degradation controller performance mismatch. In this paper, a (MPC) algorithm Gaussian process regression (GPR) proposed. Firstly, kinematic equations mobile established by mechanistic analysis method; similarly, system modeled using second-class Lagrangian equations. Secondly, problem stability reliability during operation considered. This paper uses MPC with main plus residual solve closed-loop strategy. The state at each moment decomposed into predicted first-principles state. learned GPR real-time used compensate for deviations between real model. proposed method requires fewer data samples, enhancing technique’s practicality. Finally, simulation results show that more stable achieves desired tracking faster. Compared algorithm, arrival time reduced 28% speed error controlled within 0.07.