RINGS: A Deep Reinforcement Learning Enabled Large-scale UAV Network with Distributed Navigation, Mobility Control, and Resilience

由于无人机(UAVs)的重大技术进步，在过去的几年里，无人机在民用和商业应用方面取得了爆炸性的增长。为了确保这些应用中高效可靠的自动导航和规划，迫切需要开发基础技术，使无人机具有强大的传感、通信和车载计算能力，能够快速适应动态变化的环境，对极端环境条件具有弹性，并能够安全地抵御数据污染和恶意攻击。拟议研究的跨学科性质将为不同的学生群体提供宝贵的研究机会和实践项目。该项目的目标是利用并显著推进下一代无线通信、深度机器学习、硬件感知模型生成以及稳健和值得信赖的人工智能方面的最新突破，以实现智能和弹性无人机导航和规划系统的设计。更具体地说，这个项目将开发：(A)通过多模式数据融合和机器学习辅助的实时通信辅助环境感知，用于全球状态跟踪的快速处理；(B)具有高度可扩展的计算和灵活的延迟容忍的多代理分散强化学习(RL)框架；(C)用于高效通信的基于深度学习的消息传递和用于高效星载计算的强大的硬件感知神经体系结构搜索；以及(D)全面的健壮性和安全性设计，用于保护系统免受离群点数据、恶意中毒攻击和RL系统攻击。该项目还将进行广泛的绩效评估，以验证开发的方法和算法。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Thanks to the significant technological advances in unmanned aerial vehicles (UAVs), the past few years have witnessed an explosive growth of UAVs in civilian and commercial applications. To ensure efficient and reliable auto-navigation and planning in these applications, there is an urgent need to develop the foundational technology that enables UAVs to have strong sensing, communications, and on-board computing capabilities, to adapt rapidly to the dynamically changing environment, to be resilient to extreme environmental conditions, and to be secure against data contamination and malicious attacks. The interdisciplinary nature of the proposed research will provide valuable research opportunities and hands-on projects for a diverse group of students.The goal of this project is to leverage and significantly advance the recent breakthroughs in NextG wireless communications, deep machine learning, hardware-aware model generation, and robust and trustworthy artificial intelligence, to enable the design of an intelligent and resilient UAV navigation and planning system. More specifically, this project will develop: (a) real-time communication assisted ambient sensing with multi-modality data fusion and machine learning assisted fast processing for global state tracking; (b) a multi-agent decentralized reinforcement learning (RL) framework with highly scalable computations and flexible latency tolerance; (c) deep learning based message passing for efficient communication and powerful hardware-aware neural architecture search for efficient on-board computation; and (d) comprehensive robustness and security design for system protection from outlier data, malicious poisoning attacks, and RL system attacks. The project will also conduct extensive performance evaluations to validate the developed approaches and algorithms.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.