Design of an adaptive MPC control system for unmanned ground vehicle based on FP-ADMM and RBFNN

Abstract

To address the challenges of nonlinearity, real-time performance, and parameter uncertainty in distributed six-wheel unmanned ground vehicles (UGV) operating in complex high-speed dynamic environments, this paper proposes a method for achieving precise control in such conditions. First, a dynamic predictive model is established, considering road inclination and time-varying curvature. Next, the grey wolf optimizer (GWO) is employed to adaptively optimize the weight coefficients of the model predictive control (MPC) objective function, creating an adaptive MPC (AMPC) system that enhances overall performance, allowing it to handle various complex terrains. Additionally, the Fast Proximal Alternating Direction Method of Multipliers (FP-ADMM) is utilized to solve the quadratic programming (QP) problem, improving computational efficiency and ensuring real-time performance for UGV at high speeds. To further reduce trajectory tracking errors caused by model uncertainties and external disturbances, a compensation controller based on the radial basis function neural network (RBFNN) is introduced. This controller learns from the system’s control errors and generates real-time compensation signals, combining its output with the AMPC output to perform trajectory tracking tasks for the UGV. The stability of the proposed method is proven based on Lyapunov stability theory. Experimental results from real-world testing show that the proposed method achieved trajectory tracking accuracies of 0.08 m on off-road and 0.06 m on inclined road. Compared to traditional MPC methods, this strategy demonstrates superior tracking performance on complex high-speed terrains, with significant improvements in both real-time performance and tracking accuracy.