CPS: Medium: Collaborative Research: Automated Discovery of Data Validity for Safety-Critical Feedback Control in a Population of Connected Vehicles

我们的世界目前正经历着真实世界数据量的惊人增长，然而这些数据只在有限的时间内有用或有效。例如，在交通施工期间提供给驾驶员的绕行信息在施工任务完成后就失去了效用。本项目开发了确定数据库中积累的数据有效性的方法，以回答数据何时过期的问题？在物理世界的安全关键型应用程序中，数据有效性的知识甚至更为重要：在当前的安全关键决策中，应该信任多少过去的数据？来自附近地点的数据是否可信，能够准确地反映当地的情况和条件？回答这些基本问题将影响广泛的应用，包括交通管理、国防、天气预报等，因为数据是现代社会的普遍特征。该项目开发的方法已在安全关键场景（如在可能结冰的道路上行驶）中实施和测试了联网自动驾驶汽车的控制。这项工作具有巨大的潜力，不仅可以确保即将部署的联网自动驾驶汽车的安全性，还可以提高广泛的数据和信息密集型应用的确定性和信心。这项合作研究将支持宾夕法尼亚州立大学、巴克内尔大学和马萨诸塞大学洛厄尔分校的研究生和本科生研究人员的发展。该项目还包括面向中学生的以科学、技术、工程和数学（STEM）为重点的外展活动，以扩大在网络物理系统领域的参与。该项目的研究目标是创建方法来确定数据的有效性如何随着时间的推移而衰减，以及随着距离数据收集地的距离增加而衰减。该研究是在安全关键系统的背景下进行的，即在可能结冰的道路上行驶的联网自动驾驶汽车（cav）车队，其中安全关键的道路摩擦信息通过无线数据链路共享到一个中央时空数据库，该数据库负责协调数据的平均。这些数据用于估计道路摩擦系数（即冰的存在），并传输给附近的其他联网车辆。使用Allan方差分析评估数据库中数据信任的持续时间和摩擦估计的有效性，使数据库能够内部建模和监控数据的及时性和质量。研究人员还研究了耦合的快慢反馈回路的性能指标（例如稳定性），其中快慢反馈回路在车辆层面起作用，通过数据库介导的摩擦测量预览来确保自动驾驶汽车在结冰条件下的安全运行。慢循环是空间，多车辆数据平均数据库中使用当前测量车队提供的自动驾驶汽车。然后在具有大时空范围的道路网络上运行的CAV车队的背景下检查这些功能。当在CAV上下文中实现时，这些方法可以用于任何从空间、时间或时空数据流中合成可操作信息的应用程序。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。

Our world is currently experiencing an incredible increase in the amount of real-world data available, yet that data remains useful or valid only for a finite period of time. For example, detour information provided to drivers during traffic construction loses its utility upon completion of the construction assignment. This project develops methods to determine the validity of data accumulated in databases, to answer the question: when do data expire? Knowledge of data validity is even more important in the context of safety-critical applications in the physical world: how much of the past data should be trusted to make safety-critical decisions in the present? Can data from nearby locations be trusted to accurately reflect local context and conditions? Answering these fundamental questions will impact a wide-range of applications, including traffic management, national defense, weather forecasting, etc., since data is a universal feature of modern society. The methods developed in this project are implemented and tested for control of connected autonomous vehicles in safety-critical scenarios such as driving on potentially icy roads. This work has significant potential to not only ensure safety in the imminent deployment of connected autonomous vehicles, but also improve certainty and confidence in a wide range of data- and information- intensive applications. This collaborative research will support development of graduate and undergraduate researchers at Penn State University, Bucknell University, and the University of Massachusetts Lowell. The project also includes Science, Technology, Engineering, and Math (STEM)-focused outreach activities for middle-school students to broaden participation within the field of cyber-physical systems. The research objective of the project is to create methods to determine how the validity of data decays over time, and over increasing distances away from where the data was collected. The research is conducted in the context of safety-critical systems, namely fleets of connected autonomous vehicles (CAVs) driving on potentially icy roads, where safety-critical road friction information is shared via a wireless data link to a central spatiotemporal database that mediates data averaging. This data is used to estimate the roadway friction coefficient (i.e. the presence of ice) and is transmitted to other connected vehicles in the vicinity. The time duration of data trust and validity of the friction estimates within the database are evaluated using Allan variance analysis, enabling the database to internally model and monitor data timeliness and quality. The investigators also study performance metrics (e.g., stability) of the coupled fast and slow feedback loops, where the fast loop acts at the vehicle level to ensure safe CAV operations in icy conditions using database-mediated preview of friction measurements. The slow loop is the spatial, multi-vehicle data averaging in the database using current measurements provided by a fleet of CAVs. These functionalities are then examined in the context of CAV fleets operating on road networks with large spatiotemporal extents. While implemented in a CAV context, these methods can be used in any application that synthesizes actionable information from spatial, temporal, or spatiotemporal data streams.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.