Estimation of driver control limitations and adaptation of intervention strategies based on crash data

Abstract

Sweden’s Vision Zero initiative aims to eliminate fatalities and serious injuries in road transportation. A key component is the development of Advanced Driver Assistance Systems (ADAS), supporting drivers in crash-imminent situations. However, determining the optimal timing and magnitude of system interventions remains challenging, as interventions should support rather than conflict with the driver’s intended maneuver. One proposed approach is to define driver control limitations, such as maximum braking and steering inputs, and use these limitations to support ADAS intervention design. This thesis analyzes real-world crash data to evaluate driver control limitations using four metrics: Steering Wheel Angle (SWA), Steering Wheel Angle Velocity (SWAV), Driver Deceleration Request (DDR), and Actual Time-to-Collision (ATTC). Actual vehicle deceleration was also considered relevant to distinguish between driverrelated braking demand and the resulting vehicle response. The metrics were examined across varying driving contexts, including vehicle speeds, vehicle actions, crash types, road and lighting conditions, and ADAS activation. The results showed that SWA varied slightly with maneuver type, while SWAV remained more consistent and may represent a more stable upper limit of steering behavior. Most drivers reached maximum deceleration request; however, the ATTC results showed that braking was often initiated too late to prevent a collision. DDR alone was insufficient in low-friction conditions, highlighting the importance of considering actual vehicle deceleration. ADAS activation generally had limited influence. Simulation results indicate that refined steering thresholds may allow drivers to avoid obstacles where current ADAS would intervene prematurely. These findings support incorporating driver control limitations into future ADAS development.