FMitF: Collaborative Research: Track I: Predictive Online Safety Analysis from Multi-hop State Estimates for High-autonomy on Highways

该项目的目标是为自动驾驶和半自动驾驶汽车带来安全保障。该方法是延长汽车预测其行驶路径的时间，并与周围车辆共享该路径。有了这些扩展的预测，就有可能根据车辆的设计模型来估计车辆当前和未来的行为。目前，在线正式的安全分析可以保证和监督，但过于保守的方法可能导致不良驾驶。这与使用试驾数据和机器学习来构建驾驶模型形成对比，后者很难分析。该项目旨在通过计算(1)从传感器数据中估计自动驾驶车辆的当前状态及其多跳环境，(2)预测未来4-6秒的车辆轨迹，以及(3)检查模型和预测来发现正确的平衡-所有这些都在毫秒内完成。除了开设一门关于自主性的新本科课程外，还将开设一个新的科学研讨会来探索自主性方面的类似问题。该项目旨在提供(1)基于车辆到基础设施/车辆（V2X）系统的新型传感器融合算法，(2)用于驾驶预测的首个开放的、机器可解释的智能体模型库，(3)用于模型识别的算法，以及(4)用于在线安全检查的算法。这些模块将集成到端到端系统OmniVisor中，并在密歇根大学Mcity设施的真实车辆中进行实际事故多发场景评估。这项研究将在形式化方法、混合动力系统、估计和检测理论以及移动网络等学科之间建立联系。如果成功，OmniVisor将为混合自动驾驶场景下的车辆获得安全保证提供科学依据，并在真实车辆的道路上进行实验验证。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。

The goal of this project is to bring safety assurance to autonomous and semi-autonomous vehicles. The approach is to lengthen the time that a car can predict its driving path, and share this path with surrounding vehicles. With these expanded predictions, it is possible to estimate the current and future behaviors of vehicles, according to their design models. Currently, online formal safety analysis can promise guarantees and oversight, but overly conservative approaches can lead to bad driving. This is in contrast to the use of test-driving data and machine learning to build driving models, which are difficult to analyze. The project aims to discover the right balance by computationally (1) estimating the current state of the autonomous vehicle and its multi-hop environment from sensor data, (2) predicting vehicle trajectories 4-6 seconds into future, and (3) checking the models and predictions---all in milliseconds. A new scientific workshop will be created to explore similar issues in autonomy, in addition to a new undergraduate course on autonomy.The project aims to deliver (1) new sensor-fusion algorithms over Vehicle-to-Infrastructure/Vehicle (V2X) systems, (2) a first-of-its-kind open, machine-interpretable library of agent models for driving predictions, (3) algorithms for model identification, and (4) algorithms for checking safety online. These modules will be integrated in an end-to-end system --- OmniVisor --- and evaluated in realistic accident-prone scenarios with real vehicles in University of Michigan's Mcity facility. The research will build connections across the disciplines of formal methods, hybrid dynamical systems, estimation and detection theory, and mobile networking. If successful OmniVisor will provide a scientific basis for obtaining safety assurances for vehicles in mixed-autonomy scenarios and experimentally demonstrate the approach on the road with real vehicles.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.