系统仿真学报 ›› 2025, Vol. 37 ›› Issue (7): 1649-1664.doi: 10.16182/j.issn1004731x.joss.25-0511

• 特约综述 • 上一篇    

自动驾驶仿真测试技术驱动汽车产业智能跃迁

吴建平1,2,3, 李冠洲1, 赵帅4, 黄玲5   

  1. 1.清华大学 土木工程系,北京 100084
    2.深圳清华大学研究院 无人驾驶与智能交通技术研发中心,广东 深圳 518000
    3.天府永兴实验室 零碳交通研究中心,四川 成都 610213
    4.中汽智联技术有限公司,天津 300300
    5.华南理工大学 土木与交通学院,广东 广州 510640
  • 收稿日期:2025-06-04 修回日期:2025-06-11 出版日期:2025-07-18 发布日期:2025-07-30
  • 通讯作者: 李冠洲
  • 第一作者简介:吴建平 1957年生,清华大学教授,清华大学-剑桥大学-麻省理工学院“未来交通”研究中心主任,深圳清华大学研究院无人驾驶与智能交通研发中心主任,四川天府永兴实验室零碳交通中心主任,俄罗斯工程院外籍院士,国家特聘专家,教育部“长江学者”特聘教授。主要研究领域:1) 自动驾驶与智能交通,2) 交通建模与交通仿真,3) 智慧城市与智慧交通。获国际国内省部级以上科研成果奖20余项,拥有国际国内发明专利20余项,在国际学术期刊上发表论文350多篇,在牛津大学、剑桥大学、麻省理工学院、加州大学等著名高校以及重要国际会议上做过30多场特邀演讲。吴教授是中国科协第十届全国委员会委员,世界工程组织(WFEO)工程环境委员会委员,英国工程技术学会会士(FIET),中国仿真学会首届会士,中国仿真学会常务理事,中国仿真学会自动驾驶与仿真测试专委会主任,中国智能交通协会道路车辆专委会副主任,北京、杭州、南宁、海口等城市顾问。
    吴建平(1957-),男,教授,博士,研究方向为交通行为、交通模型与交通仿真;智能驾驶与未来交通;数字孪生、城市计算与城市仿真;智慧生态城市与绿色交通等。
  • 基金资助:
    重庆市自然科学基金创新发展联合基金(长安汽车)(CSTB2024NSCQ-LZX0159);智能汽车安全技术全国重点实验室开放基金(IVSTSKL-202432)

Intelligent Transition of Automotive Industry Driven by Autonomous Driving Simulation Testing Technology

Wu Jianping1,2,3, Li Guanzhou1, Zhao Shuai4, Huang Ling5   

  1. 1.School of Civil Engineering, Tsinghua University, Beijing 100084, China
    2.Research Center for Autonomous Driving and Intelligent Transport Systems, Research Institute of Tsinghua University in Shenzhen, Shenzhen 518000, China
    3.Zero Carbon Transportation Research Center, Tianfu Yongxing Laboratory, Chengdu 610213, China
    4.CATARC Intelligent and Connected Technology Co. , Ltd. , Tianjin 300300, China
    5.School of Civil Engineering & Transportation, South China University of Technology, Guangzhou 510640, China
  • Received:2025-06-04 Revised:2025-06-11 Online:2025-07-18 Published:2025-07-30
  • Contact: Li Guanzhou

摘要:

自动驾驶仿真测试技术作为支撑智能驾驶系统安全验证与商业落地的核心手段,在技术方法与应用场景上取得了显著进展。传统真实道路测试因成本高昂、极端场景覆盖率低及效率瓶颈等问题,难以满足当前日益迫切的高等级自动驾驶(L4级及以上)的安全验证需求。仿真测试技术应包含数学建模、虚拟场景、硬件在环(HIL)、混合现实与云仿真集群等多层次的验证体系。数学建模可加速算法开发,虚拟场景仿真提升了感知系统的鲁棒性,HIL测试保障了控制器可靠性,而云仿真集群通过大规模并行计算实现了场景覆盖的指数级扩展。FLOWSIM平台基于模糊数学理论,在真实人类驾驶行为数据的基础上建立了“基因级”人类驾驶行为模型,保障了仿真测试场景中交通流环境的精确性。FLOWSIM-MR基于数字孪生实现了一种自动驾驶虚实结合的测试范式。未来随着自动驾驶技术及相应的测试技术发展成熟,新技术如生成式AI与数字孪生技术将推动仿真测试向更高精度与智能化演进,而国际标准(如ISO 34502)的制定与政产学研协同生态的构建,将成为突破“安全-成本-效率”三角困境的关键。

关键词: 自动驾驶仿真测试, 虚实结合仿真, 测试场景库构建, 自动驾驶安全验证, 驾驶行为建模

Abstract:

As a pivotal approach supporting the safety verification and commercial implementation of intelligent driving systems, autonomous driving simulation testing has achieved remarkable progress in technical methodologies and application scenarios. Conventional real-world road testing faces critical limitations including prohibitive costs, inadequate coverage of corner case scenarios, and efficiency bottlenecks, rendering it insufficient for safety validation of high-level autonomous driving systems (L4 and above). To address these challenges, simulation testing frameworks have evolved into a multi-layered verification system encompassing mathematical modeling, virtual scenarios, hardware-in-the-loop (HIL), mixed reality, and cloud-based simulation clusters. Specifically, mathematical modeling accelerates algorithm development; virtual scenario simulation enhances the robustness of perception systems, and HIL testing ensures controller reliability. Meanwhile, cloud-based simulation clusters achieve exponential expansion in scenario coverage through large-scale parallel computing. Notably, the FLOWSIM platform leverages fuzzy logic to establish a "genetic-level" human driving behavior model based on real-world driving data, ensuring the accuracy of traffic flow environments in simulated test scenarios. Furthermore, FLOWSIM-MR introduces a virtual-real testing paradigm for autonomous driving based on digital twins. Looking ahead, the maturation of autonomous driving technologies and their testing technologies will be propelled by emerging innovations like generative AI and digital twin systems, driving simulation testing toward higher precision and intelligence. Concurrently, the establishment of international standards (e.g., ISO 34502) and collaborative ecosystems involving governments, academia, and industry will be critical to overcoming the "safety-cost-efficiency" trilemma.

Key words: autonomous driving simulation testing, virtual-real simulation, testing scenario database construction, safety validation of autonomous driving, driving behavior modeling

中图分类号: