Black Icy Road Classification Enabled by Vehicle Self-Aligning Torque Estimation From Electric Power Steering Systems and 1-D CNNs

초록

This study presents a preliminary investigation of a new strategy for detecting and classifying black icy road conditions based on self-aligning torque (SAT) estimation from electric power steering (EPS) systems. Existing studies on black ice detection have largely focused on computer-vision-based approaches and on-board sensor measurements for slip ratio estimation. However, these methods often suffer from limited robustness under variations in driving maneuverability and stability, vehicle vibrations, and road surface roughness, which can lead to degraded detection accuracy and leave substantial room for improvement. Motivated by these limitations, we exploit the observation that transient vibrations in tire lateral forces and the resulting aligning moments induce pronounced transient responses in the EPS system during transitions from black ice surfaces to normal roads under active steering control. Based on this observation, we propose a Kalman filter with unknown input that augments a conventional Kalman filter with disturbance-observer capability to estimate the SAT using measured steering torque and steering angle. Features extracted from the estimated SAT are then used as inputs to a black icy road classifier based on one-dimensional convolutional neural networks. The proposed detection and classification pipeline is validated through in-vehicle experiments, demonstrating its practical effectiveness. © 2026 The Authors.

키워드