S&AS: INT: Autonomous Multi-Robot Visual Monitoring for Urban, Agricultural, and Natural Resource Management

该项目开发了改进自然资源监测系统的技术。具体地说，这个项目调查了配备视觉传感器的智能和自主空中机器人团队如何在扩展的空间和时间尺度上监测植被健康。它解决了以下基本问题：i)多机器人协作，ii)根据收集的视觉数据的质量自适应机器人轨迹，iii)多个冲突约束下的网络决策，以及iv)真实世界动态环境中的学习规划和控制。该项目进一步研究了这四个任务如何在自主和实时操作下相互无缝工作。除了改进植被监测外，该项目的成果还可能惠及其他几个应用领域，包括生境监测、边界控制、森林火灾跟踪和搜救行动。此外，该项目涉及几个动手操作程序，预计将吸引本科生参与，并能够接触到K-12学生。这项研究解决了实现对关键自然资源的自主多机器人视觉监控的理论和技术挑战。这项工作的关键一步是引入新的算法框架，使智能和自主空中机器人网络能够检测和适应观察到的现象，同时满足多个-往往相互冲突的-时变约束和任务规范。为了实现这一目标，该项目在三个主要领域做出了贡献。I)自适应视觉感知，重点是分析能够自动适应环境和计算约束的视觉数据。Ii)自主空中机器人导航，在资源受限的情况下开发可靠、节能的自主空中机器人导航策略。3)系统级的决策和适应，重点是实时约束下的自适应资源分配的一般推理、决策和控制方案。该项目将个人捐款无缝地并入拟议的自适应资源监测(ARM)系统。ARM认识到其环境、技术和道德限制，并为每个代理人发布以目标为导向的任务。ARM具有反思过去决策和强化有利决策的能力，并由丰富的知识模型和方法驱动。通过严格的评估计划对各个组件和整个系统进行测试，从各个模块开始，直到在应用程序领域进行系统级评估。

This project develops technologies to improve natural resource monitoring systems. Specifically, this project investigates how teams of smart and autonomous aerial robots equipped with visual sensors can monitor vegetation health over extended spatial and temporal scales. It addresses the fundamental issues of i) multi-robot collaboration, ii) adaptation of robot trajectories according to the quality of collected visual data, iii) networked decision making under several conflicting constraints, and iv) learning planning and control at real-world dynamic environments. The project further investigates how these four tasks can work seamlessly with each other, under autonomous and real-time operation. In addition to improving vegetation monitoring, the outcomes of this project may benefit several other application domains, including habitat monitoring, border control, forest fire tracking and search-and-rescue operations. Additionally, the project involves several hands-on procedures that are expected to attract undergraduate student participation and enable outreach to K-12 students.This research addresses the theoretical and technical challenges in order to enable autonomous multi-robot visual monitoring of critical natural resources. A critical step in this effort is to introduce novel algorithmic frameworks that can enable networks of smart and autonomous aerial robots to detect and adapt to observed phenomena, while satisfying multiple - often conflicting - time-varying constraints and mission specifications. To achieve this, the project contributes to three main areas. i) Adaptive visual sensing with focus on the analysis of visual data that can automatically adapt to the environment and the computational constraints. ii) Autonomous aerial robot navigation, developing reliable and energy-efficient autonomous aerial robot navigation strategies under resource constraints. iii) System-level decision-making and adaptation, which will focus on general reasoning, decision-making and control schemes for adaptive resource allocation under the real-time constraints. The project seamlessly merges individual contributions into the proposed Adaptive Resource Monitoring (ARM) system. ARM is cognizant of its environmental, technical and ethical constraints, and issues goal-oriented tasks for each agent. ARM possesses the ability to reflect upon past decisions and reinforce advantageous ones, and is driven by knowledge-rich models and methods. The individual components and the overall system are tested through a rigorous evaluation plan, starting with individual modules and leading up to a system-level evaluation in the application domain.