NRI: FND: Consistent distributed visual-inertial estimation and perception for cooperative unmanned aerial vehicles

NRI：FND：协作无人机的一致分布式视觉惯性估计和感知

• 批准号: 1924897
• 负责人: Guoquan Huang
• 金额: $ 38万
• 依托单位: University of Delaware
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2023-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1924897&HistoricalAwards=false
• 关键词: NRI; FND; Consistent; distributed; visual

This project is in response to the emerging demand for ubiquitous deployment of autonomous robots in real-world applications. In particular, thanks to their small size, agile maneuverability, and low-altitude flight ability even in complex environments, unmanned aerial vehicles (UAVs) have witnessed significant progress over the last decade. The ubiquitous availability of small and inexpensive UAVs that are equipped with sensing, processing, and communication capabilities, will make it possible to deploy them in teams that can collaborate to accomplish missions more efficiently and robustly than a single vehicle. Assisted by technological advances in sensing, computing, communication, and hardware design and manufacturing, in the coming years, cooperative UAVs will become valuable tools in critical applications ranging from environmental monitoring and emergency response to precision agriculture. However, when developing cooperative UAV systems, many challenges remain, among which, one of the biggest is the stringent resource limitations (such as limited computation power, communication bandwidth, and energy) that UAVs are faced with. Performing cooperative estimation and perception under resource constraints, incurs many challenges that must be addressed during the UAV operations as well as during the design of UAV systems. In this project, the investigators will design scalable, robust and distributed state estimation and 3D perception for cooperative UAVs using visual and inertial measurements under computation and communication constraints, thus providing 3D scene understanding and spatial cognition to support intelligent decision making. To this end, resource-adaptive consistent visual-inertial estimation will be formulated as constrained optimization to optimally utilize available resources. Leveraging deep learning/AI techniques, the project team will design deep neural networks to power visual-inertial 3D perception in order to semantically and spatially understand environments. To achieve optimal performance for given resources or determine cost-effective system design for desired performance, the project team will develop formal tools for characterization and co-design of UAV hardware and software systems. By technologically enabling ubiquitous deployment of UAVs, the results of this project will foster innovative applications in robotics such as aerial transportation during humanitarian aid and disaster relief, thus boosting economic development. Moreover, this project will promote hands-on learning in undergraduate education in mechanical engineering and enrich graduate curriculum in robotics, as well as create opportunities for students to perform meaningful research.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.

该项目是为了响应在现实世界中普遍部署自主机器人的新兴需求。特别是，由于其体积小，灵活的机动性和即使在复杂环境中的低空飞行能力，无人机（UAV）在过去十年中取得了重大进展。配备有传感、处理和通信能力的小型廉价无人机无处不在，这将使它们有可能部署在团队中，这些团队可以合作，比单一车辆更有效、更强大地完成任务。在传感、计算、通信以及硬件设计和制造方面的技术进步的帮助下，未来几年，合作无人机将成为从环境监测和应急响应到精准农业等关键应用中的宝贵工具。然而，在发展无人机协同系统时，还存在许多挑战，其中最大的挑战之一是无人机面临的严格的资源限制（如有限的计算能力、通信带宽和能量）。在资源受限的情况下进行协同估计和感知，会带来许多挑战，这些挑战必须在无人机操作期间以及在无人机系统的设计期间解决。在该项目中，研究人员将在计算和通信约束下使用视觉和惯性测量为合作无人机设计可扩展，鲁棒和分布式状态估计和3D感知，从而提供3D场景理解和空间认知，以支持智能决策。为此，资源自适应一致的视觉惯性估计将制定为约束优化，以最佳地利用可用资源。利用深度学习/人工智能技术，项目团队将设计深度神经网络来支持视觉惯性3D感知，以便在语义和空间上理解环境。为了在给定的资源下实现最佳性能，或者确定具有成本效益的系统设计，以实现所需的性能，项目团队将开发用于无人机硬件和软件系统的表征和协同设计的正式工具。通过在技术上实现无人机的无处不在部署，该项目的成果将促进机器人技术的创新应用，例如人道主义援助和救灾期间的空中运输，从而促进经济发展。此外，该项目将促进机械工程本科教育的实践学习，丰富机器人研究生课程，并为学生进行有意义的研究创造机会。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Fast Monocular Visual-Inertial Initialization Leveraging Learned Single-View Depth

利用学习的单视图深度进行快速单目视觉惯性初始化

• 发表时间: 2023
• 期刊: Robotics: Science and Systems (RSS
• 作者: Merrill, N.;Geneva, P.;Katragadda, S.;Chen, C.;Huang, G.
• 通讯作者: Huang, G.

Online IMU Intrinsic Calibration: Is It Necessary?

• DOI: 10.15607/rss.2020.xvi.026
• 发表时间: 2020-07
• 期刊: Robotics: Science and Systems XVI
• 作者: Yulin Yang;Patrick Geneva;Xingxing Zuo;G. Huang

Optimization-Based VINS: Consistency, Marginalization, and FEJ

• DOI: 10.1109/iros55552.2023.10341637
• 发表时间: 2023-10
• 期刊: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
• 作者: Chuchu Chen;Patrick Geneva;Yuxiang Peng;W. Lee;Guoquan Huang

Map-based Visual-Inertial Localization: A Numerical Study

• DOI: 10.1109/icra46639.2022.9811829
• 发表时间: 2022-05
• 期刊: 2022 International Conference on Robotics and Automation (ICRA)
• 作者: Patrick Geneva;G. Huang

Online Self-Calibration for Visual-Inertial Navigation: Models, Analysis, and Degeneracy

• DOI: 10.1109/tro.2023.3275878
• 发表时间: 2023-10
• 期刊: IEEE Transactions on Robotics
• 影响因子: 7.8
• 作者: Yulin Yang;Patrick Geneva;Xingxing Zuo;G. Huang