Tsoutsanis A., Ziakopoulos A., Yannis G., “Self-Supervised Detection of Harsh Cornering Events at Scale using Smartphone Sensor Data”, Proceedings of the Transport Research Arena TRA 2026 Conference, Budapest, 18-21 May 2026.

This study presents a self-supervised framework for detecting harsh cornering events using smartphone telematics under real-world deployment conditions. GPS-derived trajectory features, orientation-invariant inertial magnitudes, and OpenStreetMap validation are combined to robustly identify candidate turns despite arbitrary device placement and sensor noise. High-confidence pseudo-labels are generated via ensemble anomaly detection, enabling supervised classifiers to learn without manual annotation. Among four evaluated models, a multi-layer perceptron achieved the best performance, and its latent embeddings revealed clear separation between harsh and normal cornering. The results demonstrate that consumer-grade smartphones can reliably capture lateral risky maneuvers at scale, extending prior work focused mainly on longitudinal events such as hard braking. This approach offers a scalable foundation for applications in driver risk assessment, insurance telematics, fleet management, and urban safety planning.