CRII: CNS: Supporting Resilient Perception in Autonomous Cyber-physical Systems

包括自动驾驶汽车、环境监测无人机和搜救机器人在内的自主网络物理系统（A-CPS）有望在没有人类帮助的情况下导航具有挑战性的物理景观。为了确保任务成功，了解可能导致A-CPS设备故障的内部或外部条件非常重要。最先进的技术是在部署前验证A-CPS可以在恶劣条件下运行，然后在部署后进行持续监测。然而，很难预测A-CPS设备一旦部署到现实世界中可能遇到的所有困难，如果在设备运行时观察到故障情况，通常为时已晚。该项目的重点是提前预测A-CPS的故障，并采取措施避免故障的发生。特别是，该项目分析了在商用嵌入式系统上执行的最先进的对象检测应用程序，使用机器学习技术生成数据驱动的故障预测模型，并实现了一个软件管理器来配置系统以避免预测故障。该项目的成功将为物理环境与A-CPS行为之间的关系提供新的见解，并将使开发人员能够以更高的信心创建更可靠、更有弹性的算法。自主的网络物理系统可以在没有人为干预的情况下导航物理世界，可以显著提高生活质量和安全。例如，仅加州就有超过3100万英亩的荒地，从北部到南部边界都有高或非常高的火灾危险严重区域。自动野火探测无人机可以帮助拯救可能成为野火受害者的居民和负责监测高风险地区的公务员的生命。为了取得成功，这种系统必须能够处理困难和不可预测的环境条件。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。

Autonomous cyber-physical systems (A-CPS), including self-driving vehicles, environmental monitoring drones, and search-and-rescue robots, are expected to navigate challenging physical landscapes without human assistance. To ensure mission success, it is important to understand what internal or external conditions may cause A-CPS devices to fail. The state-of-the-art is to verify that A-CPS can operate in adverse conditions before deployment, followed by continuous monitoring post-deployment. However, it is difficult to anticipate all of the difficulties an A-CPS device may encounter once deployed in the real world, and if failure conditions are observed while the device is operating, it is typically too late to react. This project focuses on predicting failures for A-CPS ahead of time and injecting actions to avoid the failures. In particular, the project analyzes a state-of-the-art object detection application executing on a commercial-of-the-shelf embedded system, generates data-driven predictive models of failure using machine learning techniques, and implements a software manager to configure the system to avoid predicted failures. The success of the project will provide new insights into the relationship between physical environments and A-CPS behaviors and will enable developers to create more reliable and resilient algorithms with higher confidence.Autonomous cyber-physical systems that can navigate the physical world without human intervention could significantly improve quality of life and safety. For instance, California alone has over 31 million acres of wildlands, with high or very high fire hazard severity zones, spanning from the northern to the southern borders. Autonomous wildfire detection drones could help save lives of residents that are potential wildfire victims, as well as public servants responsible for monitoring high risk regions. To be successful, such systems must be able to handle difficult and unpredictable environmental conditions.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.