汽车工程 ›› 2023, Vol. 45 ›› Issue (8): 1428-1437.doi: 10.19562/j.chinasae.qcgc.2023.08.013

所属专题: 智能网联汽车技术专题-规划&决策2023年

• • 上一篇    下一篇

基于错误注入的决策规划系统抗扰性测试与分析

吴新政,邢星宇,刘力豪,沈勇,陈君毅()   

  1. 同济大学汽车学院,上海 201804
  • 收稿日期:2023-02-01 修回日期:2023-03-17 出版日期:2023-08-25 发布日期:2023-08-17
  • 通讯作者: 陈君毅 E-mail:chenjunyi@tongji.edu.cn
  • 基金资助:
    国家重点研发计划(2022YFB2503001);国家自然科学基金重点项目(52232015);重庆市技术创新与应用发展专项重大主题专项项目(cstc2019iscx-zdztzxX0041)

Testing and Analysis of the Robustness of Decision-Making and Planning Systems Based on Fault Injection

Xinzheng Wu,Xingyu Xing,Lihao Liu,Yong Shen,Junyi Chen()   

  1. School of Automotive Studies,Tongji University,Shanghai 201804
  • Received:2023-02-01 Revised:2023-03-17 Online:2023-08-25 Published:2023-08-17
  • Contact: Junyi Chen E-mail:chenjunyi@tongji.edu.cn

摘要:

自动驾驶系统的运行环境复杂多样。考虑到传感器本身的性能局限及感知算法在特定触发条件下的功能不足,自动驾驶系统上游感知结果不可避免地会出现错误。因此针对自动驾驶决策规划系统在上游数据错误情况下的抗扰性测试对保证自动驾驶安全性至关重要。为此,本文首先提出基于6层场景本体模型的数据模型和包含4类不确定性错误模式的错误模型,并进一步构建了一个通用的错误注入框架SOFIF以实现对上游数据的修改。最后,本文基于硬件在环仿真测试并提出危害率作为量化评价指标,对比分析了两个被测决策规划系统在存在不确定性错误模式下的抗扰性表现。实验得到两个被测系统的危害率分别为0.89和0.64,表明两被测系统的抗扰性存在较大差距,并进一步证明了SOFIF的有效性。

关键词: 自动驾驶, 预期功能安全, 测试与评价, 抗扰性测试, 错误注入

Abstract:

Automated driving systems operate in complex and diverse environments. Considering the performance limitations of the sensors and the functional insufficiency of the perception algorithms under certain trigger conditions, it is inevitable that the upstream perception results of the autonomous driving system will be incorrect. Therefore, it is essential to test the robustness of decision-making and planning systems under conditions of erroneous upstream data to ensure the safety of automated driving. Firstly, in this paper, a data model based on a six-layer scenario ontology model and a fault model containing four types of uncertainty error patterns are proposed. Further, a generic fault injection framework named SOFIF is constructed to enable modification of upstream data. Finally, the robustness of two decision-making and planning systems under the error patterns of uncertainty existence is compared and analyzed based on Hardware-in-the-Loop (HiL) simulation testing, with the hazard rate proposed as the quantitative evaluation index. The hazard rate of the two tested systems is 0.89 and 0.64, respectively, indicating a large gap in the robustness of the two tested systems and further proving the effectiveness of SOFIF.

Key words: automated driving, safety of the intended functionality, testing and evaluation, robustness testing, fault injection