NRI: Collaborative Research: Autonomous Quadrotors for 3D Modeling and Inspection of Outdoor Infrastructure

该项目开发的技术用于收集视觉和惯性数据，为桥梁、发电厂和炼油厂等民用和工业基础设施构建离线高精度三维结构地图。它还开发了在线处理技术，包括定位、路径规划和避障。该项目建立了一个使用四旋翼机的系统，以协助他们的人类同事对室外基础设施进行视觉检查，以提高此类操作的效率和效果。该研究在传感、估计和控制的关键领域推进了当前的艺术状态，使小型四旋翼机能够协助人类进行视觉检查。除了提高国家基础设施的可靠性外，该项目还通过为四旋翼飞行器开发开源模块化算法，使机器人领域的研究人员、开发人员、教育工作者和最终用户受益。该项目提供与当地努力和全州倡议相一致的教育和社区外展活动，并通过与当地高中的合作，寻求增加多样性，吸引代表性不足的群体学习科学、技术、工程和数学（STEM）。这项研究解决了来自传感和处理限制的基本挑战，这些限制阻碍了在视觉检测任务中使用低成本、小尺寸的四旋翼飞行器。它侧重于一个四步骤的过程，其中最初的四旋翼是远程操作在一个安全的距离，从感兴趣的结构收集必要的视觉和惯性数据，离线，高精度的三维地图的结构。然后，检查工程师使用这些地图来指定感兴趣的区域。最后，四旋翼飞行器利用机载传感器对结构进行精确定位，并沿着检查路线导航，同时收集额外的数据，以提高未来检查的准确性和可靠性。一个关键的创新是使信息以多种形式和抽象层次提供，以满足离线（例如，检查区域的可视化和规划信息丰富的路径）和在线（例如，基于地图的定位和避障）使用的经常相互冲突的需求。同样关键的是信息驱动的方法，以最大限度地利用有限的传感和处理资源提供给四旋翼。最后，该方法的一个关键优点是，它为每次检测飞行后的精度和效率的持续改进提供了基础。

This project develops technologies to collect visual and inertial data necessary for constructing, offline, high-accuracy 3D maps of the structure for civil and industrial infrastructure such as bridges, power plants, and refineries. It also develops technologies for online processing including localization, path planning and obstacle avoidance. The project builds a system that employs quadrotors to assist their human co-workers in visual inspections of the outdoor infrastructure to enhance efficiency and effectiveness of such operations. The research advances the current state of the art in key areas of sensing, estimation, and control necessary for enabling small-size quadrotors to assist humans in visual inspections. In addition to improving the reliability of the nation's infrastructure, the project benefits researchers, developers, educators, and end-users in robotics by developing open-source, modular algorithms for quadrotors. The project offers educational and community outreach activities aligned with local efforts and state-wide initiatives, and seeks to increase diversity and attract underrepresented groups to Science, Technology, Engineering, and Mathematics (STEM) via a partnership with local high schools. This research addresses the fundamental challenges stemming from sensing and processing limitations that prevent the use of low-cost, small-size quadrotors in visual-inspection tasks. It focuses on a four-step process, where initially a quadrotor is tele-operated at a safe distance from the structure of interest to collect visual and inertial data necessary for constructing, offline, high-accuracy 3D maps of the structure. These maps are then used, by the inspection engineer, to designate areas of interest. Lastly, the quadrotor employs its onboard sensors to precisely localize with respect to the structure and navigate along the inspection route, while collecting additional data for increasing the accuracy and improving the reliability of future inspections. A key innovation is making information available in multiple forms and levels of abstraction so as to meet the often-conflicting needs of offline (e.g., visualization of inspection areas and planning information-rich paths) and online (e.g., map-based localization and obstacle avoidance) uses. Also critical is an information-driven approach for making maximum use of the limited sensing and processing resources available to the quadrotor. Lastly, a key advantage of the proposed approach is that it provides the foundation for continual improvement in accuracy and efficiency after each inspection flight.