Lee, Eunho (Ajou University), YOO, SUNGMIN (Ajou Universtiy), Park, Joonho (Ajou University), So, Jaehyun (Ajou University)

Identification and Classification of Adverse Road Traffic Features Impacting Autonomous Vehicles under Naturalistic Driving Conditions

Scheduled for presentation during the Regular Session "S42b-Safety and Risk Assessment for Autonomous Driving Systems" (FR-EA-T42), Friday, November 21, 2025, 14:50−14:50, Broadbeach 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Takeover Requests (TORs) are critical control transitions in Level 2–3 autonomous driving systems, typically triggered when environmental complexity surpasses system capabilities. While previous studies have predominantly focused on internal vehicle triggers such as driver state or sensor degradation, this study investigates TOR behavior through the lens of environmental conditions using empirical data. A total of 538 TOR events were collected from autonomous shuttle operations in Anyang, South Korea. Thirteen contextual features—capturing aspects of road geometry, visibility, and traffic flow—were used to perform K-means++ clustering, resulting in four distinct environmental profiles. To quantify risk, an exposure-adjusted metric was developed, normalizing TOR frequency by the occurrence probability of each environmental context. This revealed disproportionate TOR risk in nighttime, high-curvature segments. In parallel, TOR events were categorized into five scenario types via rule-based classification, with high-complexity cases such as Localization Drift predominantly emerging in visually and geometrically demanding conditions. The findings highlight the need for environment-aware TOR logic and adaptive HMI design in real-world deployments. The proposed framework offers a scalable approach for TOR risk modeling based on contextual profiles. Future work includes integration of rule-based and probabilistic triggering models and scenario-driven simulation for response evaluation.