CRII: NeTS: Towards Predictive Communications for UAV based IoT Networks

2016年，全球无人机出货量超过200万架，创造了45亿美元的全球市场收入，预计这一数字还将进一步增加，在未来智能互联社区中，无人机将具有改变现代生活的巨大潜力。无人机将通过实现新的应用，并简化现有的应用，包括运输、交通控制、远程健康监测、监视、边境巡逻、栖息地监测和精准农业，显著影响我们的日常生活。在这些领域，基于无人机的解决方案商业化的一个重要缺点是网络效率低下。目前的通信协议在适应无人机网络的动态拓扑结构方面效率极低。该项目旨在解决这个问题，并通过预测网络拓扑变化来开发可预测的通信策略。因此，该研究将通过促进高通量信息交换和消除对多个地面控制站的需求，为完全自主的无人机网络开发认知通信协议。该项目旨在为具有不同机动性水平的飞行物网络开发新的预测通信工具。关键思想是在预期的网络分区和/或故障之前通过通信协议采取预防措施。提出的方法依赖于开发一个通用模型来预测周围网络节点的运动轨迹。为了实现运动分析，将每个对象的驱动力建模为具有相同类型对象共享的隐藏层的分层生成模型。使用新颖的合并和分割方法提供了灵活性，可以容纳具有看不见的可操作性级别的新对象类，并撤销入侵对象。本文将利用范畴论中基于循环模群的新方法来设计网络拓扑预测的最佳测量模式。最后，在寻找最优路径时，将预测的网络拓扑结合到决策中，开发一种预测路由算法。本文将基于随机矩阵浓度不等式对路径最优性和预测不确定性之间的权衡进行具体表征，为实际实现铺平道路。将研究图快照器和混合路由技术，以整合传统和预测路由算法，以加速大规模网络路由算法的执行。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。

More than 2 million Unmanned Aerial Vehicles (UAVs) have shipped in 2016 creating a global market revenue of $4.5 billion, and is projected to increase further, with huge potential to transform modern living in future smart & connected communities. UAVs will significantly impact our daily life by enabling new applications, and simplifying existing applications including transportation, traffic control, remote health monitoring, surveillance, border patrolling, habitat monitoring, and precision agriculture. An important drawback to commercialize UAV-based solutions in these domains is networking inefficiency. Current communication protocols are extremely inefficient in accommodating dynamic topologies of networks of UAVs. This project seeks to address this issue and develop predictive communication strategies by anticipating network topology changes. The research will thus develop cognizant communication protocols for fully autonomous UAV networks by facilitating high-throughput information exchange and eliminating the need for multiple ground control stations. This project aims to develop novel tools for predictive communication for networks of flying objects with heterogeneous maneuverability levels. The key idea is to take preventive actions via communication protocols before anticipated network partitions and/or failures. The proposed methodology relies on developing a universal model to predict motion trajectories of surrounding network nodes. To enable motion profiling, driving force of each object is modeled as a hierarchical generative model with a hidden layer shared among objects of the same type. The use of a novel merge-and-split method provides flexibility in accommodating new object classes with unseen maneuverability levels and repealing intruding objects. Novel methods based on cyclic monoids in category theory will be used to design optimal measurement patterns for network topology prediction. Finally, a predictive routing algorithm will be developed by incorporating the predicted network topology into decision making when looking for the optimal path. A concrete characterization of the tradeoff between the routing optimality and prediction uncertainties will be developed based on random matrix concentration inequalities to pave the road for practical implementation. Graph snappers and hybrid routing techniques will be studied to integrate conventional and predictive routing algorithms to accelerate the routing algorithm execution for large-scale networks.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.