Collaborative Research: NRI: INT: Cooperative Underwater Structure Inspection and Mapping

该项目开发了一个与人类操作员合作的协作机器人系统，以绘制水下结构图。水下结构图是一项重要的能力，适用于海洋考古、基础设施维护、资源利用、安全和环境监测等多个领域。水下环境对人类来说在许多方面都是具有挑战性和危险的，而机器人操作与水面上的操作相比面临着更多的挑战。特别是，传感和通信都受到限制，需要在信息有限的情况下进行三维规划。该项目将生成一个水下结构的3D模型，提供高分辨率的照片真实感表示。自主水下机器人(AUV)将密切合作，生成基于视觉的密集观测表面重建，并与远程人类操作员协调。该项目通过培训本科生和研究生将研究和教育结合在一起，他们将有机会在南卡罗来纳州、新泽西州和新罕布夏州的一个包容的跨学科团队中工作。该系统将被整合起来，并在实地进行考古测绘。研究将沿着三个方向进行。(1)基于深度学习的稳健水下状态估计方法和三维重建的混合表示，该表示将对导航和用户初始检查的概率占用进行编码。(2)协同规划，基于局部优化框架，考虑信息增益、不确定性减少、环路闭合、远端观察者主动定位、用户偏好联合测量和通知近端观察者去哪里等多个准则，对近端观察者进行协同规划。(3)信息驱动的通信，精心设计了3-D重建的有效数据表示和决定何时以及如何共享的跨层优化。这三个部分将有助于实现使协作机器人团队能够自主操作并生成水下结构的真实地图的总体目标。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。

This project develops a system of co-robots collaborating with a human operator to map underwater structures. Underwater structure mapping is an important capability applicable to multiple domains: marine archaeology, infrastructure maintenance, resource utilization, security, and environmental monitoring. The underwater environment is challenging and dangerous for humans in many aspects, while robotic operations face additional challenges compared to the above-water ones. In particular, both sensing and communications are restricted, and planning is required in three dimensions with limited information. The project will generate a 3D model of the underwater structure providing a high-resolution photo-realistic representation. Autonomous Underwater Vehicles (AUVs)will be operating in close cooperation, generating a dense vision-based reconstruction of the observed surface, and coordinated with remote human operators.. The project integrates research and education through training of undergraduate and graduate students, who will have the opportunity to work in an inclusive, interdisciplinary team across South Carolina, New Jersey, and New Hampshire. The system will be integrated and tested for archaeological mapping at field sites. Research will be conducted along three directions. (1) Robust underwater state estimation based on a deep learning approach and a hybrid representation for 3-D reconstruction that will encode probabilistic occupancy for both navigation and initial inspection from users. (2) Collaborative planning, for the proximal observers based on a local optimization framework that originally considers multiple criteria, including information gain, uncertainty reduction, and loop closure, active positioning of distal observers, and user preference to make joint measurements and inform proximal observers on where to go. (3) Information driven communications, with careful design of efficient data representation of the 3-D reconstruction and of a cross-layer optimization for deciding when and how to share. These three components will contribute towards the overarching goal of enabling a team of co-robots to operate autonomously and produce a realistic map of an underwater structure.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.