Abstract:

A quantitative safety-driving evaluation framework was proposed for autonomous driving functions of intelligent connected vehicles (ICVs) in China. Grounded in a standardized functional assessment protocol, the framework specifically targeted two critical safety-related capabilities, which were “identification and response to the dynamic states of surrounding vehicles”, and “identification and response to pedestrians and non-motorized vehicles”. A comprehensive evaluation model was developed, integrating three methodological components, which were extraction of expert driver behavioral features from naturalistic driving data, vehicle inverse dynamics decoupling to isolate control-relevant motion states, and F-norm-based matrix quantification of trajectory deviation and response timeliness. A high-fidelity co-simulation environment was constructed to enable rigorous validation by integrating PreScan, CarSim and Simulink. The quantitative safety-driving scores were obtained by applying this framework to two ICV platforms equipped with an industry-standard black-box autonomous driving system. The results demonstrate that the proposed method yields scores significantly more consistent with empirically observed expert driving behavior. Relative to conventional evaluation approaches, the framework improves overall performance assessment accuracy by 38.11% and 68.57% for the two test vehicles, respectively.

Key words: automated driving general detection, driving safety, empirical driving, F-norms matrix, quantitative evaluation