SaTC: CORE: Small: Online Malicious Intent Inference for Safe CPS Operations under Cyber-attacks

现代自动驾驶汽车在制造时并没有考虑到安全问题。增加的传感、计算、控制能力和任务复杂性已经引入了超越传统网络攻击的安全问题。通过注入错误的数据，通过欺骗传感器，通过篡改控制器，甚至通过操纵环境，攻击者可以破坏这些网络物理系统的完整性，甚至控制其功能。最近在各种系统上展示了攻击的例子，包括改变无人机的路线，通过全球定位系统（GPS）欺骗劫持船只，以及使用无线连接来接管汽车的转向和刹车。近年来，已经有几种解决方案被用于解决这个问题，然而，大部分网络物理系统安全文献都集中在恶意攻击的检测和估计上，而没有考虑上下文、风险或后果，更不用说攻击的意图了。相比之下，预测攻击意图可以获得更多关于攻击的信息，从而提供防御机制。本研究的重点是识别、预测和减轻自动驾驶汽车恶意意图的技术开发，寻求开发评估恶意攻击风险和后果的基本方法，识别恶意意图，以及防御、控制和重新配置受损系统。该项目将为提高自动驾驶汽车的弹性提供基本方法。具体而言，建议的研究包括：1)利用系统模型和基于可达性分析的知识来评估攻击风险和后果的新技术；2)基于机器学习和控制级意图推理方法；3)制定弹性规划和控制政策，确保系统在闭环性能保障下持续运行。为了更好地开发和评估本工作中提出的安全技术，将使用具有不同传感、计算和通信能力的最先进的无人驾驶空中和地面车辆实施现实案例研究，以促进其向实践的过渡。所提出的研究也广泛适用于网络物理系统，并将直接促进安全自主系统的发展。此外，作为该项目的一部分，重点将放在教育和推广上，包括开发以机器人弹性为主题的新课程活动，让本科生和研究生参与研究，以及与工业界合作，培训下一代网络物理系统安全问题和缓解方案的劳动力。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。

Modern autonomous vehicles are not built with security in mind. The increased sensing, computation, control capabilities, and task complexity have introduced security concerns beyond traditional cyber-attacks. By injecting malformed data, by spoofing sensors, by tampering with controllers, and even by manipulating the environment, an attacker can compromise the integrity and even take control over the functionality of such cyber-physical systems. Examples of attacks have recently been demonstrated on a variety of systems which include the rerouting of drones, the hijacking of vessels by Global Positioning System (GPS) spoofing, and the use of wireless connectivity to take over the steering and brakes of automobiles. Several solutions have been pursued in recent years to solve this problem, yet the bulk of cyber-physical system security literature is focused on the detection and estimation of malicious attacks without considering the context, risk or consequences, much less the intent of the attack. Predicting the intention, by contrast, may yield more information about the attack and thus offer defense mechanisms. This research focuses on the development of techniques to identify, predict, and mitigate malicious intentions of autonomous vehicles, seeking to develop fundamental methods for estimating risk and consequences of malicious attacks, identifying malicious intent, and defending, controlling, and reconfiguring the compromised system.This project will provide fundamental approaches to increase resiliency in autonomous vehicles. Specifically, the proposed research includes: 1) new techniques to estimate risk and consequences of attacks, leveraging knowledge about the system model and reachability-based analysis; 2) machine learning-based and control-level intent inference methods; and 3) the development of policies for resilient planning and control to ensure continuous operation of the system with closed-loop performance guarantees. To better develop and assess the security techniques proposed in this work, realistic case studies will be implemented using state-of-the-art unmanned aerial and ground vehicles with different sensing, computation, and communication capabilities to facilitate their transition into practice. The proposed research is also applicable to cyber-physical systems broadly and will contribute directly to the development of safe autonomous systems. Additionally, as part of this project, a major emphasis will be given to education and outreach including the development of novel curriculum activities centered on the topic of resiliency in robotics, involvement of undergraduate and graduate students in research, and collaborations with industry to train the next generation workforce on cyber-physical system security problems and mitigation schemes.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.