S&AS: INT: Traffic Deconfliction for Smart and Autonomous Unmanned Aircraft Systems in Congested Environments

无人机（UAV）的操作带来了可观的科学和社会效益，其应用范围从包裹递送，线性基础设施检查到大规模环境监测。然而，联邦航空管理局（FAA）还不允许在没有空中追逐飞机或地面观察员的情况下自主UAV操作以确保交通冲突，这是复杂且昂贵的。拟议项目的目标是使给定的无人机能够预测其相邻无人机的轨迹，动态评估碰撞风险，并自动解决潜在的冲突，而无需持续的人工监督。该项目旨在：1）实现以可能满足未来FAA法规的方式在没有直接人为监督的情况下操作多个UAV的能力; 2）建立UAV交通的数学表示，其可以被缩放以处理复杂、异构和高维交通;以及3）成熟的无人机操作，用于民用和商用，与现有的商业空中交通相协调。该项目旨在开发一个真实的-时间，分布式决策算法，基于动态网络的概率路径规划器（DNPRP），用于在拥挤的环境中多个无人机的自主操作。关键技术组件将包括1）用于无人机数据共享的云数据库系统，2）虚拟和动态无人机网络发现算法，3）基于时间依赖的贝叶斯非参数高斯过程回归轨迹预测模型，以及4）用于无冲突路线生成的时间依赖碰撞概率图。对于物理飞行测试，DNPRP将通过使用在封闭的室外设施（即，弗吉尼亚理工大学无人机公园）数小时至数天，没有人为干预。实时操作中的计算成本和解决方案最优性之间的权衡将严格评估可测量的指标，这将通知可能的扩展到大规模复杂系统。通过与行业合作伙伴的密切合作，将研究成果快速转化为开源或商业软件。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。

Unmanned air vehicle (UAV) operations bring considerable scientific and societal benefits with its applications ranging from package delivery, linear infrastructure inspection, to large-scale environmental monitoring. However, Autonomous UAV operation is not yet allowed by the Federal Aviation Administration (FAA) without airborne chase aircraft or ground observers to ensure traffic deconfliction, which is complicated and costly. The goal of the proposed project is to enable a given UAV to predict the trajectories of its neighboring UAVs, dynamically assess collision risks, and automatically resolve potential conflicts without the need for constant human supervision. This project seeks to: 1) implement the ability to operate a multitude of UAVs without direct human oversight in a way that might satisfy future FAA regulations; 2) establish a mathematical representation of a UAV traffic that can be scaled to handle a complex, heterogeneous, and high-dimensional traffic; and 3) mature UAV operations for civil and commercial applications in harmony with existing commercial air traffic.The project aims at developing a real-time, distributed decision-making algorithm, the Dynamic Network-based Probabilistic Route Planner (DNPRP), for autonomous operation of multiple UAVs in congested environments. The key technical components will include 1) a cloud-sourced database system for UAV data sharing, 2) an algorithm for virtual and dynamic UAV network discovery, 3) time-dependent, Bayesian nonparametric Gaussian process regression-based trajectory prediction models, and 4) a time-dependent collision probability map for conflict-free route generation. For physical flight testing, DNPRP will be evaluated by using a couple of multirotor UAVs that operate within an enclosed outdoor facility (i.e., the Virginia Tech Drone Park) for hours to days without human intervention. Trade-offs between computational cost and solution optimality in real-time operation will be assessed rigorously with measurable metrics, which will inform possible extensions to large-scale complicated systems. A rapid path for transitioning the research results to open-source or commercial software will be formed in close collaboration with an industrial partner.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.