Data-Driven Characterization of Motorcycle Riders’ Kinematics and Crash Risk

Abstract

This effort was successful in exploring different motorcycle riding styles using multiple, commonly available sensors, ultimately associating those results with potential crash events. More specifically, the analysis carried out in this work provides novel insights into real-world motorcyclist behavior by identifying three distinct riding profiles characterized by unique kinematic patterns. Furthermore, several potential kinematic indicators that may predict crash risk were identified. This enhanced characterization of motorcycle rider capabilities could enable more realistic crash scenario simulations, inform evidence based safety policies, and support the design of advanced rider assistance systems that leverage real-world parameters. In this study, riding styles and their relationship with crash and near-crash (CNC) risks were investigated using naturalistic riding study data from 155 participants over an average period of 11 months. The data, predominantly from southern California, includes approximately 400,000 miles of riding, providing extensive insight into real-world motorcycle riding behaviors. The research addresses two main questions:

What defines the normal riding behavior of a motorcycle? Can it be categorized in terms of riding style? What measures of rider performance are associated with crash risk? Kinematic data collected from the instrumented motorcycles was processed to remove artifacts and noise prior to analysis. The analytical approach identified three distinct riding style clusters through principal component analysis and K-means clustering. The first cluster of 24 participants was primarily composed of younger participants using sport motorcycles. These subjects exhibited higher accelerations, abrupt braking, and substantially higher roll angles more frequently than other clusters. Additionally, this group showed significantly higher CNC rates than other clustered riders. Cluster 2 exhibited typical riding behavior, with moderate acceleration, braking, and roll rate values across a mix of motorcycle types and rider ages. In contrast, the riding behavior of the third cluster was smoother, with less aggressive maneuvers and safer distances during car-following scenarios. This third cluster was mainly composed of older participants primarily riding cruiser motorcycles. Statistical analysis revealed that age and motorcycle type significantly differentiated these clusters, whereas gender, riding experience, and formal training had minimal impact. A bootstrap analysis comparing CNC trips against trips without a CNC (i.e., baseline trips) identified abrupt initial jerk during braking and accelerating maneuvers as a significant indicator of elevated crash risk, underlining the critical role of anticipatory and reactive behaviors in rider safety. These findings offer critical insights for targeted rider education, motorcycle design, safety policy formation, and evaluation and design of advanced rider assistance systems. This enhanced understanding of motorcycle kinematic metrics and their linkage with crash risk can also inform insurers as they develop more precise, data-driven telematics-based rate structures.