Estimation of road adhesion coefficient using an innovation-based adaptive extended Kalman filter for vehicle active safety

Real-time and accurate estimation of the road adhesion coefficient is pivotal for enhancing vehicle active safety systems (e.g. ABS, ESP) and ensuring reliable autonomous driving decisions. However, conventional Kalman filter-based methods (EKF, UKF) suffer from slow convergence, insufficient accuracy, and poor robustness under sudden road changes—rooted in fixed process noise assumptions and inadequate adaptation to strong nonlinear tire dynamics. To address these issues, this study proposes an innovation-based adaptive extended Kalman filter (AEKF) for real-time estimation of four-wheel independent adhesion coefficients. A 3-degree-of-freedom (3-DOF) vehicle dynamics model integrated with the Dugoff tire model is established to characterize tire forces under combined slip, forming a nonlinear state-space system with four-wheel adhesion coefficients as state variables and sensor-measurable longitudinal acceleration, lateral acceleration, and yaw rate as observations. The AEKF’s core innovation is a normalized prediction residual statistic-based adaptive mechanism, dynamically adjusting process noise covariance to enhance responsiveness and robustness. Comprehensive CarSim/Simulink co-simulations (high/low adhesion, abrupt transitions) show the AEKF outperforms EKF/UKF, reducing the average root mean square error (RMSE) by 61.94%–70.86% compared with the traditional EKF, and 36.61%–74.24% against the UKF, with convergence time shortened by over 50% under sudden changes. It provides a reliable perception basis for real-time control of active safety systems and decision-making of autonomous driving, effectively improving vehicle stability and collision avoidance capability.