CAREER: Systematic Approach for Extensively (SAfEly) Testing and Verifying the Security of Connected and Autonomous Vehicle

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。互联自动驾驶汽车（CAV）的潜在好处数不胜数，社会期望这项技术能够提高日常生活质量，并兑现其承诺。然而，为了有效，它们必须经过测试，以证明达到标准的安全和保障水平。运输系统的复杂性和相互联系的性质使得测试和验证的任务极其困难，引起了对其安全性的严重关切。因此，它要求新的问题形式和新的系统方法来完成CAV测试和验证的任务。现有的测试解决方案使用特定的方法，例如行驶里程，来证明一些安全指标，通常假设CAV对周围环境的感知是全面和理想的。然而，目前还没有基本的结构来证明CAV产品的安全性。CAREER建议将交通系统建模为一个网络控制系统，提供一种新的弹性度量，使自动驾驶汽车的弹性测试成为可能。此外，它利用先前开发的验证框架来制定测试和验证过程，作为集中反馈控制系统，使新型攻击发生器的开发成为可能。该项目的预期结果将为自动驾驶汽车的安全测试铺平道路，直接影响这项技术和相关标准的未来，最终消除与车祸相关的死亡事故，挽救生命。研究结果可以进一步应用于所有网络控制系统，例如高保证军事系统和从无人机到电力系统的自主系统。该项目的教育目的是通过设计充分整合的教育模块和示范，扩大学生，特别是代表性不足的少数民族和妇女对CAV安全的认识。我们计划包括以下活动，以满足农村和大部分经济贫困地区的需求：(i)开发针对小学，高中和大学生的课后在线STEM课程；（ii）为教育工作者和工业伙伴提供讲习班，作为他们的专业发展活动；（iii）通过研究本科生经历和实习项目，让代表性不足的本科生和大学生参与进来；（四）开设本科和高级研究生课程。这个CAREER项目解决了自动驾驶汽车安全性的测试和验证问题。现有文献已经认识到自动驾驶汽车安全性的重要性，并推动了几种检测和补偿算法的开发，以确保在故障、故障和攻击下的安全性。然而，在cav测试和验证任务上投入的精力并不多。这个CAREER项目表明，目前的方法不足以在现实环境中安全地验证自动驾驶汽车的安全性，因为缺乏动态依赖的度量来衡量系统的弹性。我们描述了一个研究计划，其中交通系统被建模为一个网络控制系统，其中道路，行人，车辆和交通标志（由于它们的动态行为）被建模为代理，使用传感器和通信网络相互交互。新的视角允许我们提出一种新的弹性度量，与安全度量一起用于开发基于强化学习的控制器，以测试自动驾驶汽车的安全性。由于故障和攻击的类型是无限的，因此所提出的控制器可以制定攻击的影响，而不是专注于特定类型，从而简化了故障和攻击产生的过程。该项目预计将通过以下方式推进自动驾驶汽车的测试和验证领域：(i)使用网络控制系统的概念引入一种新颖的视角，通过将测试过程建模为反馈控制系统，利用强化学习开发独特的数据流生成器，从而产生攻击，其中最小化安全性是期望的目标；（ii）开发一个独特的实验平台，丰富了混合现实（MR）的力量。以及车辆在环（vehicle-in- loop, ViL），以安全测试自动驾驶汽车的安全性。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).The potential benefits of connected autonomous vehicles (CAV) are numerous, and society is expecting that this technology will increase the quality of everyday life and follow through on its promises. However, to be effective, they must be tested to demonstrate a standard level of safety and security. The complex and interconnected nature of the transportation system makes the task of testing and verification exceedingly difficult, raising serious concerns regarding their safety and security. It, thus, calls for new problem formulation and a novel systematic approach for the task of CAV testing and verification. The existing testing solutions use ad-hoc methods, such as miles driven, to demonstrate some indication of safety, often assuming that the CAV's perception of the surrounding environment is comprehensive and ideal. However, no fundamental structure has been developed to demonstrate the security of CAV products. This CAREER proposal models the transportation system as a networked control system providing a novel resiliency metric enabling the testing resiliency of CAVs. In addition, it utilizes the prior developed verification framework to formulate the testing and verification process as a centralized feedback control system enabling the development of a novel attack generator. The expected outcomes of this project would pave the way towards safely testing CAVs, directly impacting the future of this technology and related standards, ultimately eliminating crash-related fatalities and saving lives. The research findings can be further implemented for all networked control systems, such as high-assurance military systems and autonomous systems ranging from unmanned aerial vehicles to power systems. The educational purpose of the project is to expand students', particularly underrepresented and women minorities, awareness of CAV security by designing fully integrated educational modules and demonstrations. We plan to include the following activities to serve the need for rural and largely economically distressed regions: (i) develop after school online STEM curriculum adjusted for primary, High-school, and college students; (ii) provide workshops for educators and industrial partners as their professional development activities; (iii) involve underrepresented undergraduate and college students through research for undergraduate experience and internship program; (iv) develop an undergraduate and an advanced graduate courses.This CAREER project addresses the problem of testing and verification for the security of CAVs. The importance of the security of CAVs has been recognized in the existing literature and has motivated the development of several detection and compensation algorithms to ensure safety under faults, failures, and attacks. However, not much effort is invested in the task of CAVs testing and verification. This CAREER project illustrates that the current approaches are insufficient to safely verify the security of CAVs in a realistic environment, suffering from the lack of a metric that is dynamic-dependent to measure the system resiliency. We describe a research plan where a transportation system is modeled as a networked control system where roads, pedestrians, vehicles, and traffic signs (due to their dynamic behavior) are modeled as agents, interacting with each other using sensors and communication networks. The new perspective allows us to propose a novel resiliency metric to be used alongside the safety metric to develop reinforcement learning-based controllers for testing CAVs' security. As there are infinite types of faults and attacks, the proposed controller formulates the effects of attacks rather than focusing on specific types, easing the process of fault and attack generation. This project is expected to advance the area of testing and verification of CAVs by (i) Introducing a novel perspective using the concept of networked control systems enabling the development of a unique data stream generator utilizing reinforcement learning to generate attacks by modeling the testing process as a feedback control system where minimizing safety and security is the desired objective and (ii) Developing a unique experimental platform enriched with the power of mixed reality (MR) and vehicle-in-the-loop (ViL) to test the security of CAVs safely.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.