NANO-Based Tire-Road Friction Estimation with Uncertainty-Conditioned Vision-Dynamics Fusion

Abstract

Accurate estimation of the tire-road friction coefficient (TRFC) is essential for vehicle active safety systems including ABS and ESC. However, purely model-based estimators suffer from response lag during abrupt surface transitions, while purely vision-based estimators are vulnerable to per-frame noise. To address these limitations, we propose a vision-dynamics fusion framework for robust and real-time TRFC estimation. Firstly, on the dynamic side, we apply the Natural Gradient Gaussian Approximation (NANO) filter to the four-wheel cascaded estimation problem and compare it with classical Cubature Kalman Filter (CKF) and Square-Root CKF (SRCKF) baselines. Secondly, we design a vision-dynamics fusion module, termed UCAF, which integrates the dynamic estimate with the visual stream through an uncertainty-conditioned weighting scheme and enforces a one-sided downward correction by construction, providing a conservative bound on the fused output. Finally, the framework is evaluated on a CarSim-Simulink co-simulation platform under a multi-segment road surface profile. Results show that NANO achieves lower tracking error and shorter settling delay than the CKF and SRCKF baselines, and that UCAF preserves the conservative bound throughout abrupt friction transitions.