【摘要】 面向自动驾驶车辆预期功能安全（SOTIF）场景的不同测试标定要求和侧重，本文提出了一种基于行车安全场（DSF）理论的SOTIF场景风险评估方法。首先，利用DSF对场景的各层元素进行风险量化，从而实现风险的集成计算。通过分析SOTIF场景的定义与架构和DSF模型的参数，证明该模型满足SOTIF场景的风险评估要求。接着将所提方法应用于3类车辆运行场景的划分中，分别是已知安全、已知不安全和未知安全/不安全。为实现场景的划分，将DSF理论中不同的驾驶状态与SOTIF中车辆的运行场景进行匹配。最后，进行了封闭场地和开放道路的测试。一方面将相对驾驶安全系数指标RDSI与碰撞时间TTC指标作对比，验证了RDSI可更准确、敏感地评估行车风险。另一方面，证明了所提方法可有效地实现场景划分。更多还原

【Abstract】 Facing different test calibration requirements and emphases of the safety of the intended functionality(SOTIF)scene for autonomous vehicles(AVs),a risk assessment method of SOTIF scene based on driving safety field(DSF)theory is proposed in this paper. Firstly,DSF is utilized to conduct risk quantification on each layer of scene elements,and hence to achieve integrated risk calculation. The definition,architecture of SOTIF scene,and the parameters of DSF model are analyzed to prove that DSF model meets the risk assessment requirements of SOTIF scene. Then,the method proposed is applied to divide the vehicle operation scenes into three types:known safe,known unsafe and unknown safe/unsafe. For realizing scene division,different driving states in DSF theory are matched with the operational scenes of vehicles in SOTIF. Finally,closed field tests and road tests are carried out. On one hand,the relative driving safety indicator(RDSI)is compared with time-to-collision(TTC)to verify that RDSI can more accurately and sensitively assess the driving risk. On the other hand,it is proved that the method proposed can effectively fulfill scene division.更多还原