弹道
毒物控制
计算机科学
运输工程
工程类
模拟
人为因素与人体工程学
医疗急救
医学
物理
天文
作者
Peixing Zhang,Bing Zhu,Jian Zhao,Tianxin Fan,Yuhang Sun
标识
DOI:10.1016/j.aap.2022.106926
摘要
Automated driving technology has constantly been maturing; however, how to ensure automated vehicle (AV) safety has not yet been effectively solved, functional safety assessment remains an important part of the development of automated driving technology. To compensate for the lack of multidimensional evaluation indicators, this paper proposes a safety evaluation method in multi-logical scenarios (SEMMS) for AVs' functional safety based on naturalistic driving trajectory (NDT) in order to evaluate the comprehensive performance of the tested AV in a diversity of scenarios simultaneously. The potential field method is used to describe the quantified danger level of an AV in a single concrete scenario that considers the dangerous situation of the scenario and AV test results. Combined with the internal probability distribution of the logical scenario parameter space obtained by NDT, the safety performance of an AV in logical scenario is calculated by integrating the two indexes. With the information entropy and relative frequency of different logical scenarios, the relative weights of logical scenarios are obtained, and the safety performance evaluation results of the tested AV in the multi-logical scenarios can be determined based on the weighting danger level in different logical scenarios. During the actual application of the method, the HighD database was used as the input source of NDT, and a black-box automated driving algorithm was subjected to traversal tests in three logical scenarios. The test results of the automated driving algorithm were evaluated using the SEMMS, and the results show that the SEMMS could well evaluate the performance of the tested automated driving algorithm in multiple kinds of logical scenarios simultaneously, indicating that it is an effective solution to the problem of automated driving algorithm safety evaluation.
科研通智能强力驱动
Strongly Powered by AbleSci AI