Adaptive Model Predictive Controller for a 4WS4WD Autonomous Ground Vehicle

Abstract

The increasing demand for autonomous ground vehicles (AGVs) necessitates advanced control strategies to ensure precise path tracking under varying conditions. This study addresses the challenge of controlling a four-wheel steer and four-wheel drive (4WS4WD) AGV to follow complex trajectories accurately, despite dynamic city driving road conditions and uncertainties like wheel slip. The research aims to develop an adaptive Model Predictive Controller (MPC) that optimizes steering angles and driving forces in real time for enhanced maneuverability and stability. A dynamic 4WS4WD vehicle model was formulated, incorporating longitudinal, lateral, and yaw dynamics, and updated online using sensor data (e.g., GPS, LiDAR, IMUs) to adapt to changing parameters like friction and velocity. The methodology involved MATLAB/Simulink simulations across eight scenarios, combining dry/wet tarmac, constant/varying velocities, and single/double lane changes, using a sedan with a 1060 kg mass and a 24-step prediction horizon. Results demonstrate superior tracking on dry tarmac with constant velocity max lateral error 0.03 m, max yaw error 0.14 rads, while wet tarmac and varying velocities have lateral error at 0.06 m and max yaw error of 0.19 rads, all within safe limits. The adaptive MPC outperforms traditional methods, reducing errors by up to 25% compared to prior benchmarks. These findings enhance AGV reliability for real-world applications, with implications for improving safety and efficiency in autonomous driving systems.