Recently, Tavakoli et al.proposed a self-testing scheme in the prepare-and-measure scenario, showing that self-testing is not necessarily based on entanglement and violation of a Bell inequality [Phys.Rev.A 98 062307(...Recently, Tavakoli et al.proposed a self-testing scheme in the prepare-and-measure scenario, showing that self-testing is not necessarily based on entanglement and violation of a Bell inequality [Phys.Rev.A 98 062307(2018)].They realized the self-testing of preparations and measurements in an N → 1(N ≥ 2) random access code(RAC), and provided robustness bounds in a 2 → 1 RAC.Since all N → 1 RACs with shared randomness are combinations of 2 → 1 and 3 → 1 RACs, the3 → 1 RAC is just as important as the 2 → 1 RAC.In this paper, we find a set of preparations and measurements in the3 → 1 RAC, and use them to complete the robustness self-testing analysis in the prepare-and-measure scenario.The method is robust to small but inevitable experimental errors.展开更多
为构建科学合理的城市道路自动驾驶测试场景,基于美国加利福尼亚州机动车管理局(California Department of Motor Vehicles, DMV)2021—2023年公开的280起自动驾驶汽车(Autonomous Vehicle, AV)碰撞事故报告,挖掘典型危险场景并完成测...为构建科学合理的城市道路自动驾驶测试场景,基于美国加利福尼亚州机动车管理局(California Department of Motor Vehicles, DMV)2021—2023年公开的280起自动驾驶汽车(Autonomous Vehicle, AV)碰撞事故报告,挖掘典型危险场景并完成测试场景转化。首先,通过多元Logistic回归分析提取人员受伤情况的显著影响因素。其次,引入独热编码(One-Hot Encoding)对分类变量进行二进制向量转换,消除传统标签编码的数值顺序偏差。然后,采用二阶聚类算法挖掘典型危险场景组,并进一步通过交叉表分析场景组与事故结果变量、道路环境变量间的关联性。最后,将危险场景转化设计为自动驾驶测试场景。结果显示,独热编码处理后的变量,聚类质量较传统方法提升50%;聚类分析共识别出12类典型危险场景,且交叉表分析表明场景组与事故结果及道路环境变量显著相关;进一步结合事故机理与测试需求,将这12类危险场景归纳为6类代表性测试场景,其中“AV停止或减速状态下被后方直行车辆追尾”的场景最为典型,在全部场景中占比46.1%。研究表明,独热编码方法显著提升了聚类分析的准确性,基于真实事故数据的场景聚类方法能识别AV在城市道路的事故模式,并为自动驾驶测试场景库的优先级划分与标准化设计提供数据驱动支撑。展开更多
基金Project supported by the National Natural Science Foundation of China(Grant Nos.61572081,61672110,and 61671082)
文摘Recently, Tavakoli et al.proposed a self-testing scheme in the prepare-and-measure scenario, showing that self-testing is not necessarily based on entanglement and violation of a Bell inequality [Phys.Rev.A 98 062307(2018)].They realized the self-testing of preparations and measurements in an N → 1(N ≥ 2) random access code(RAC), and provided robustness bounds in a 2 → 1 RAC.Since all N → 1 RACs with shared randomness are combinations of 2 → 1 and 3 → 1 RACs, the3 → 1 RAC is just as important as the 2 → 1 RAC.In this paper, we find a set of preparations and measurements in the3 → 1 RAC, and use them to complete the robustness self-testing analysis in the prepare-and-measure scenario.The method is robust to small but inevitable experimental errors.
文摘为构建科学合理的城市道路自动驾驶测试场景,基于美国加利福尼亚州机动车管理局(California Department of Motor Vehicles, DMV)2021—2023年公开的280起自动驾驶汽车(Autonomous Vehicle, AV)碰撞事故报告,挖掘典型危险场景并完成测试场景转化。首先,通过多元Logistic回归分析提取人员受伤情况的显著影响因素。其次,引入独热编码(One-Hot Encoding)对分类变量进行二进制向量转换,消除传统标签编码的数值顺序偏差。然后,采用二阶聚类算法挖掘典型危险场景组,并进一步通过交叉表分析场景组与事故结果变量、道路环境变量间的关联性。最后,将危险场景转化设计为自动驾驶测试场景。结果显示,独热编码处理后的变量,聚类质量较传统方法提升50%;聚类分析共识别出12类典型危险场景,且交叉表分析表明场景组与事故结果及道路环境变量显著相关;进一步结合事故机理与测试需求,将这12类危险场景归纳为6类代表性测试场景,其中“AV停止或减速状态下被后方直行车辆追尾”的场景最为典型,在全部场景中占比46.1%。研究表明,独热编码方法显著提升了聚类分析的准确性,基于真实事故数据的场景聚类方法能识别AV在城市道路的事故模式,并为自动驾驶测试场景库的优先级划分与标准化设计提供数据驱动支撑。
