Collaborative Research: CISE: Large: Integrated Networking, Edge System and AI Support for Resilient and Safety-Critical Tele-Operations of Autonomous Vehicles

自动驾驶汽车（AVs）配备了一名车载人类安全驾驶员，多年来一直在公共道路上进行测试，目前有几家公司正在选定的美国城市提供“机器人出租车”试验服务。安全在运输中至关重要，因为错误可能是昂贵的，危险的或致命的。关于机器人出租车在街道上制造混乱的新闻报道凸显了复杂的现实交通环境所带来的挑战。显然，尽管人工智能（AI）和机器学习（ML）取得了快速发展，但完全自动驾驶的自动驾驶汽车还有很长的路要走。AV远程操作被建议作为一种替代方法，其中人类操作员远程控制AV，可能仅部分地根据需要出现。这一概念的灵感来自于新兴的第五代（5G）网络所提供的潜力。然而，到目前为止，用于AV远程操作的5G仍然更具抱负，因为仍然存在许多挑战。该项目的目标是解决在5G和下一代（NextG）网络上支持（部分）AV远程操作的挑战。该项目有助于促进（遥控）自动驾驶汽车的安全和增量采用，以应对社会挑战，同时加速自动驾驶技术实现完全自主。特别是，它为在中西部冬季和其他场景中测试AV远程操作提供了独特的机会。该项目也是一个论坛，为政府-行业合作和技术翻译，并作为一个连接点，扩大参与研究，教育和社区外展。这个跨学科和变革的研究议程开发集成的网络，系统和人工智能支持AV远程操作。主要创新包括：1）面向语义和细粒度的网络框架，利用多样性提供高带宽和低延迟; 2）针对人工智能工作负载优化的敏捷、安全的边缘系统架构; 3）新型应用驱动、跨层和全系统方法，实现终端设备、网络、边缘系统和人工操作员之间的协作; 4）一种“AI原生”范式，系统地集成跨层和系统组件的AI/ML算法，内置机制以减轻不准确或错误的AI预测的风险;最后，5）以人为本的方法，结合比实时更快的AI模拟和集成的机器&人类智能，无缝支持人在/在环。这些创新被整合到一个名为NextMOVE的原型AV远程操作平台中，该平台为（部分）远程操作的AV提供弹性，安全关键的支持。这些创新广泛适用于其他工业4.0用例，包括智能制造、精准农业和远程医疗，这些对国家繁荣、安全和福祉至关重要。该奖项反映了NSF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 vehicles (AVs), with an in-vehicle human safety driver, have been tested on public roads for years, and several companies are now offering “robotaxi” trial services in selected US cities. Safety is of the utmost importance in transportation, as mistakes can be expensive, dangerous, or fatal. News stories about robotaxis creating havoc on the streets highlight the challenges posed by complex real-world traffic environments. Clearly, AVs with fully autonomous driving still have a long way to go, in spite of rapid advances in artificial intelligence (AI) and machine learning (ML). AV tele-operations are suggested as an alternative approach, wherein a human operator remotely controls an AV, perhaps only partially as the need arises. This notion is inspired by the potential offered by emerging fifth-generation (5G) networks. However, as of now, 5G for AV tele-operations remains more aspirational, as many challenges remain. The goal of this project is to tackle the challenges in supporting (partial) AV tele-operations over 5G and next-generation (NextG) networks. This project helps facilitate safe and incremental adoption of (tele-operated) AVs to address societal challenges, while accelerating AV technology towards full autonomy. In particular, it provides a unique opportunity for testing AV tele-operations in Midwest winter and other scenarios. The project also serves as a forum for academia-government-industry collaboration and technology translation, and as a nexus point for broadening participation in research, education, and community outreach.This interdisciplinary and transformative research agenda develops integrated networking, systems and AI support for AV tele-operations. Key innovations include: 1) a semantics-oriented and fine-grained networking framework that exploits diversity to provide high bandwidth and low latency; 2) an agile, secure-by-design edge systems architecture that is optimized for AI workloads; 3) a novel application-driven, cross-layer and whole-system approach that enables cooperation across end devices, networks, edge systems and human operators; 4) an “AI-native” paradigm that systematically integrates AI/ML algorithms across layers and system components, with built-in mechanisms to mitigate the risks of inaccurate or false AI predictions; and finally, 5) a human-centered approach that combines faster-than-real-time AI simulations and integrated machine & human intelligence to seamlessly support human-in/on-the-loop. These innovations are incorporated into a prototype AV tele-operation platform called NextMOVE that provides resilient, safety-critical support for (partially) tele-operated AVs. The innovations broadly apply to other Industry 4.0 use cases, including smart manufacturing, precision agriculture and tele-health, which are vital to national prosperity, security, and well-being.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.