Abstract:

To clarify the safety boundaries and performance differences of the combined driving assistance system (CDAS), and solve the problems of single existing evaluation condition, insufficient scenario adaptation and unclear coupling mechanism of multiple influencing factors, a systematic evaluation methodology system for CDAS safety events was constructed, which integrated scenario classification, function decomposition and multi-dimensional quantification. Four mass-produced vehicle models were taken as research objects, and real-vehicle road evaluations were carried out in six cities. 36 types of sub-functions, 10 types of driving behaviors and 3 levels of scene complexity were defined in detail, and typical driver emergency intervention events were extracted through data slicing. With the driver emergency intervention disengagement rate and interval mileage as indicators, a comparative analysis was conducted on scene complexity and over-the-air (OTA) updates. The results show that there are differences in disengagement rates among different sub-function scenarios; the disengagement rate in high-complexity scenarios is 8.97%~19.88%, which is significantly higher than that in low-complexity scenarios (<5%); the disengagement rate in game scenarios is 3~7 times that in non-game scenarios; OTA improves the safety of most sub-function scenarios, while performance degradation is prone to occur in complex scenarios such as right turns at intersections, evasion, and merging/diverging. The safety performance of CDAS is jointly affected by the coupling of traffic scene complexity, game scenarios and OTA. It is suggested that a scenario-based priority matrix should be established in the algorithm iteration process, the training data should be balanced to avoid the bias effect of training.

Key words: combined driver assistance systems (CDAS), urban scenarios, driver emergency intervention, evaluation methodology, safety events