Simulation test can reduce time costs by accelerating testing and quickly improve test scenario coverage, especially edge scenarios with low probability of occurrence. How to reasonably and accurately define and calculate the coverage of autonomous driving (AD) simulation test scenarios is a very important issue. This article proposes a scenario enumeration method for AD behavior safety, and defines test scenario coverage from four levels, namely functional scenario theoretical coverage, functional scenario construction coverage, logical scenario construction coverage, and concrete scenario parameter coverage. The first two layers represent the entire traffic environment with typical limited functional scenarios. The last two layers represent behavioral patterns with reasonable parameter ranges. Based on the data-driven method, the completeness of coverage is proved and the adequacy of simulation verification are ensured. This research can be used to evaluate the coverage of simulation scenarios, prove the completeness of the simulation scenario database, and guide the development and testing of autonomous driving systems (ADS).

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Research on Test Scenarios Coverage for the Behaviour Safety of Autonomous Driving Vehicles

  • Yu Tang,
  • Yingzhi Xiong,
  • Junfu Huang,
  • Qiang Zhang,
  • Jinying Zhou,
  • Xiaona He,
  • Kan Yi

摘要

Simulation test can reduce time costs by accelerating testing and quickly improve test scenario coverage, especially edge scenarios with low probability of occurrence. How to reasonably and accurately define and calculate the coverage of autonomous driving (AD) simulation test scenarios is a very important issue. This article proposes a scenario enumeration method for AD behavior safety, and defines test scenario coverage from four levels, namely functional scenario theoretical coverage, functional scenario construction coverage, logical scenario construction coverage, and concrete scenario parameter coverage. The first two layers represent the entire traffic environment with typical limited functional scenarios. The last two layers represent behavioral patterns with reasonable parameter ranges. Based on the data-driven method, the completeness of coverage is proved and the adequacy of simulation verification are ensured. This research can be used to evaluate the coverage of simulation scenarios, prove the completeness of the simulation scenario database, and guide the development and testing of autonomous driving systems (ADS).