Model Predictive Rear-Wheel Assist Control for Path Tracking of Autonomous Mobility Based on Steering Performance Degradation Monitoring

This study proposes a driver monitoring and active rear-wheel assist steering control scheme integrating Model Predictive Control (MPC) and Recursive Least Squares (RLS) to enhance path-following precision and facilitate seamless control authority distribution in autonomous mobility. Rather than attempting to directly measure internal physiological cognitive states, the proposed approach quantifies physical steering performance degradation by employing a dual RLS algorithm to estimate a Steering Performance Degradation Index, which systematically fuses temporal response delay and spatial tracking deviation. Based on this real-time index, an assist MPC dynamically computes the auxiliary rear-wheel steering angle by adapting its tracking input weights according to three candidate weighting functions: exponential, linear, and threshold-based. High-fidelity co-simulations in IPG CarMaker and MATLAB/Simulink are conducted under various velocities and road curvatures with systematic driver delays. The evaluation results demonstrate that the proposed assist controller effectively enhances path-tracking precision, reducing the maximum lateral error and yaw angle error by up to approximately 82.24% and 73.53%, respectively, compared to the unassisted delayed driver. These findings verify that the proposed steering control architecture successfully mitigates transient trajectory deviation during driver performance degradation, establishing a promising candidate fail-safe strategy for advanced automated driving systems.