Title: Leveraging Machine Learning on Production Vehicle Data for Binary Road Surface Classification During Cruising

Abstract:

Accurate, real-time assessment of road slipperiness—or ideally, a precise estimate of peak grip potential—is essential for the effective operation of vehicle warning and intervention systems. Traditional approaches typically rely on dynamics-based recursive estimators that calculate the slip slope to determine friction. However, the performance of these methods is significantly limited by specific vehicle dynamic scenarios. Specifically, when a vehicle is cruising with minimal or no slip, such techniques often fail. This ineffectiveness stems from the inability of current production-grade sensors, such as wheel speed sensors, and existing methodologies to measure or accurately estimate micro slip, a factor critical for differentiating between various road surfaces.

To overcome this limitation, this study investigates the correlation between vehicle signals and road conditions during cruising through the application of machine learning. We employ both a feature-based framework and an end-to-end data-driven framework to link statistics of vehicle dynamic behavior with road surface states. The system performs binary classification, categorizing conditions as either "grip" (dry or damp) or "slip" (snow or ice).

The methodology utilizes a sliding-window approach to batch short, buffered sequences of data including wheel speeds, wheel torques, longitudinal acceleration, steering angle, and yaw rate. These inputs are then processed by a machine learning module to predict the road state. Validation using public-road data demonstrates that the data-driven approach can correctly identify road surfaces even during cruising. These findings suggest significant potential for accurate, data-driven friction-related state estimators within the domain of tire and vehicle dynamics.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC