Experimental investigation on developing human-centric AEB based on drivers' collision-avoidance capability in safety-critical scenarios.

Advanced Driving Assistance Systems (ADAS) can mitigate traffic accidents by issuing warnings or executing automated interventions in human-vehicle shared driving systems. However, their adaptability to human drivers remains limited, leading to inadequate coordination, increased collision risk, and potentially severe occupant injuries in safety-critical scenarios. This limitation stems from insufficient integration of drivers' inherent collision-avoidance behaviors into system design. Existing research lacks quantitative methods for assessing how drivers exhibit varying collision-avoidance capabilities across different automation levels. To address these gaps, we designed three types of safety-critical highway scenarios and conducted a simulator-based experiment to collect collision-avoidance data across different automation levels. Analyzing 45 drivers (mean age: 27.9 ± 5.1 years; mean driving experience: 7.8 ± 3.3 years; mean annual mileage: 6,890 km) across 2,605 scenarios, we quantified collision-avoidance capability using naturalistic behaviors and collision rates. Independent validation showed that collision risk increased by 59.7 percentage points when scenario urgency exceeded the derived collision-avoidance capability boundary. On this basis, we developed an a capability-based Autonomous Emergency Braking (AEB) concept with graded braking thresholds anchored to the empirically derived driver capability boundary. Results demonstrate that, compared with conventional AEB, the capability-based system achieves a 17.2% reduction in collision rates and mitigates average occupant injury severity by 10%. Furthermore, the system enhances operational reliability by decreasing nuisance intervention rates by 24.4%. This study demonstrates the feasibility of adapting AEB intervention timing based on driver capabilities, providing methodological insights that could inform the future design of broader human-centric ADAS and address the takeover challenge in shared driving.