Classifying interaction patterns of powered two-wheelers during severe conflicts with cars at intersections: a naturalistic driving data study

At intersections, conflicts between powered two-wheelers (PTWs) and cars are a major contributor to traffic accidents due to PTWs’ highly dynamic behavior. This study analyzes naturalistic driving data from three Chinese cities, comprising 1,871 PTW–car conflict trajectories. 16 features were used to characterize PTW behavior before, during, and after interactions. Through classification and K-means++ clustering, 9 distinct interaction patterns were identified, which were then categorized into four types: anticipative, responsive, adaptive, and proactive. PTW behaviors were further analyzed from three layers: target, tactic, and operation. Results show that PTWs maintain a consistent safety margin, with a post-encroachment time of about 3.3 ± 0.8 s; tactical decisions depend on the initial expectation of reaching the conflict point; despite with similar conditions and results, PTWs exhibit diverse actions due to individual preferences. These findings improve understanding of PTW dynamics and can enhance autonomous vehicles to better anticipate and respond to PTWs.