Defining Preventable Boundaries in Automated Driving Systems: A Driver Behavior Model for Scenario-Based Assessments

This study presents a quantitative framework for defining preventable boundaries in scenario-based assessments of Automated Driving Systems (ADS) grounded in a novel driver behavior model. This approach integrates data from real traffic and two driving simulation studies focusing on cut-in maneuvers on highways. In the first experiment, the evasive responses and decision times of drivers were statistically modeled based on real traffic data to characterize the role of cut-in responders, highlighting the behavioral differences between experienced and inexperienced drivers. Building on the first experiment findings, the second experiment analyzed the behavior of the maneuvering initiator in scenarios in which an ADS performed a cut-in maneuver, capturing reaction times, braking strategies, and risk perception. Statistical comparisons and regression analyses were used to derive preventability criteria, revealing that human drivers prioritize steering over braking in imminent-risk situations, and that experience level significantly influences response timing and choice. By combining empirical data, systematic reviews, and scenario-specific statistical modeling, this study quantifies the variability in human responses, challenging the sufficiency of the fixed-threshold safety benchmarks currently used in ADS validation. Additionally, cross-validation with real-world lane-change datasets demonstrated discrepancies with the prevailing safety models, suggesting overconservatism in the current assumptions. The results provide methodological insights for refining ADS testing standards, and contribute to data-driven regulatory guidance grounded in human performance analytics.