Road Rough Estimation for Autonomous Vehicle Based on Adaptive Unscented Kalman Filter Integrated with Minimum Model Error Criterion

The accuracy of road input identifiaction for autonomous vehicles (AVs) system, especially in state-based AVs control for improving road handling and ride comfort, is a challenging task for the intelligent transport system. Due to the high fatality rate caused by inaccurate state-based control algorithm, how to precisely and effectively acquire road rough information and chose the reasonable road-based control algorithm become a hot topic in both academia and industry. Uncertainty is unavoidable for AVs system, e.g., varying center of gravity (C.G.) of sprung mass, controllable suspension damping force or variable spring stiffness. To tackle the above mentioned, this paper develops a novel observer approach, which combines unscented Kalman filter (UKF) and Minimum Model Error (MME) theory, to optimize the estimation accuracy of the road rough for AVs system. A full-car nonlinear model and road profile model are first established. Secondly, a MME criterion is proposed to deal with the varying system parameters and model error of AVs. Then, the unscented Kalman filter based the MME theory is used to form adaptive unscented Kalman filter (AUKF) observer. Finally, compared with the traditional UKF approach, the corresponding estimation accuracy of road rough information are analyzed by using the MATLAB software and full-car test rig platform. Simulation and experimental results show that the higher accuracy of the proposed AUKF method compared with traditional UKF for AVs system improves more than 12% under the same external input condition. The research achievements develop a reasonable algorithm to apply to the road management and improving chassis performance for AVs.