基于场景的自动驾驶汽车虚拟测试研究进展

随着自动驾驶等级的提高，面向传统汽车的测试工具与测试方法已不能满足自动驾驶汽车测试的需要。基于场景的虚拟测试方法在测试效率、测试成本等方面具有巨大的技术优势，是未来自动驾驶汽车测试验证的重要手段，已成为当前的研究热点。通过对大量相关文献的系统梳理，综述了基于场景的自动驾驶汽车虚拟测试研究进展。对比分析了自动驾驶测试场景的不同定义方式，明确了测试场景的内涵，归纳了测试场景的要素种类，概述了测试场景的数据来源，总结了场景数据的处理方法。在此基础上，对自动驾驶汽车虚拟测试方法进行了总结，分析了典型的测试方式、测试平台和虚拟测试的技术要点，梳理了软件在环、硬件在环和车辆在环测试方案及其关键技术。针对自动驾驶汽车测试效率问题，研究了基于场景的加速测试技术，概述了典型的测试场景随机生成方法和危险场景强化生成方法。最后，对基于场景的自动驾驶汽车虚拟测试所面临的问题及未来发展趋势进行了分析和展望。研究结果表明：基于场景的虚拟测试是推动自动驾驶技术发展和产业落地的必由之路，未来研究应着力突破基于解构与自动重构的测试场景数据库、人-车-环境系统一体化高置信度建模、自动驾驶汽车虚拟测试标准工具链、不同自动驾驶汽车渗透率下的混合交通模拟与测试、测试案例动态自适应随机生成机制等核心共性技术，建立自动驾驶汽车虚拟测试标准体系。

Review of Scenario-based Virtual Validation Methods for Automated Vehicles

Owing to the advancement in autonomous driving technology, testing tools and testing methods for conventional automobiles cannot meet the validation requirements of autonomous vehicles. Scenario-based virtual validation methods have technical superiority with respect to testing efficiency and time consumption. Such methods can aid in conducting autopilot test verification in the future and have thus drawn significant research interest. In this study, through the systematic analysis of a large number of related literature, the developmental history of scenario-based virtual testing associated with autonomous vehicles is reviewed. The differences among scenario definitions were compared and the connotation of a test scenario was defined. Various types of elements, data sources, and processing methods associated with scenarios were specified, based on which, various virtual testing methods for autonomous vehicles were identified and listed. Typical virtual testing methods, test platforms, and virtual test points were subsequently analyzed, and key technologies corresponding to software-in-the-loop testing, hardware-in-the-loop testing, and vehicle-in-the-loop testing were outlined. To address the issue of inadequate testing efficiency associated with the virtual testing process, scenario-based automated driving acceleration test technology was studied. Typical random scenario test generation methods and crucial scenario reinforcement generation methods were thereby determined and listed. Furthermore, issues and future development trends associated with scenario-based virtual validation of autonomous vehicles were also analyzed. The obtained results indicate that scenario-based virtual testing is essential for promoting the development of autopilot technology. Additionally, further research needs to be conducted on aspects, such as the development of test scenario database based on deconstruction and automatic reconfiguration, high-confidence models for human-vehicle-environment integration system, primary technologies for supporting virtual testing of automated vehicle driving through standard tool chains, mixed traffic simulation with testing under different autonomous vehicle penetration rates, iterative optimization and adaptive acceleration testing of automated driving, and establishment of standard systems for virtual testing of autonomous vehicles.