Abstract: A vehicle-ride-accident scenario test-condition-library and the corresponding test condition set evaluation model were established to test and evaluate the safety of autopilot in vehicle rider scenarios. 116 vehicle cyclist crashes were screened out. Seven accident scenarios were extracted by using the method of classification tree based on the traffic accident database of a city in Jiangsu Province of China, and based on the National Vehicle Accident In-depth Investigation System (NAIS) database of China. An overall test condition set of the scenarios was generated according to the road and environmental parameters. An evaluation model of vehicle and cyclist crash scene is established by using an analytic hierarchy process (AHP) and a fuzzy comprehensive evaluation method. The results show that as a typical type of accident test condition in China, two-wheeled-vehicle crossing has the highest score and the highest risk on an unobstructed road section in a sunny day with both a passenger-car and a two-wheeled-vehicle driving in a same direction.

Key words: automatic driving vehicles, accident evaluation, vehicle rider scenarios, database, National Automobile Accident In-Depth Investigation System (NAIS), analytic hierarchy process (AHP)