CPS: Synergy: Tracking Fish Movement with a School of Gliding Robotic Fish

这个项目的重点是开发一种技术，通过植入声学标签来连续跟踪活鱼的运动，使用一种相对便宜的水下机器人网络，称为滑行机器鱼。这项研究解决了系统设计中的两个基本挑战：(1)适应由于环境扰动、通信延迟和鱼类运动中明显的随机性而产生的显著不确定性，以及(2)在满足船上计算、通信和电力资源的多种约束的同时，平衡相互竞争的目标(例如，机器人网络的准确跟踪和长寿命)。鱼类运动数据为选择栖息地、迁徙路线和产卵行为提供了洞察。通过推进最先进的鱼类追踪技术，该项目能够更明智地作出渔业管理和养护的决策，包括控制入侵物种、恢复本地物种和对高价值物种进行种群评估，并最终促进渔业和水生生态系统的可持续性。通过推进滑翔机器鱼网络的协调和控制，并使其能够在五大湖等具有挑战性的环境中作业，该项目还促进了这些机器人系统在环境监测、港口监视和水下结构检查中的各种其他应用的实际采用。该项目加强了密歇根州立大学的几门研究生课程，并为学生提供了独特的跨学科培训机会，包括那些来自代表性不足群体的学生。推广活动，包括机器鱼演示，博物馆展品，教师培训，以及Follow That Fish智能手机应用程序，都是专门为激发大学预科学生对科学和工程的兴趣而设计的。该项目的目标是创建一个综合框架，用于设计耦合的机器人和生物系统，以严格和整体的方式适应系统的不确定性和相互竞争的目标。这一目标是通过追求与鱼类运动跟踪和建模的应用相关的五个紧密耦合的研究目标来实现的：(1)开发新的机器人平台以实现水下通信和声学标签检测，(2)开发具有分析性能保证的健壮算法来基于多个机器人之间的到达时间差异来定位标记的鱼，(3)设计隐马尔可夫模型和在线模型自适应算法以有效和高效地捕捉鱼的运动，(4)探索用于机器人完成鱼跟踪的双层决策体系结构，其结合了鱼运动的模型预测、能量消耗和移动性约束，以及(5)对设计框架进行实验评估，首先在内陆湖泊中定位或跟踪静止和移动的标签，然后在休伦湖的桑德湾跟踪和模拟湖泊鲑鱼在产卵期间的运动。这个项目提供了对健壮的机器人-物理-生物系统的设计的基本见解，以应对系统不确定性和相互竞争的目标的挑战。首先，提出了一种用于机器人和生物组件之间紧密交互的反馈范式，以便于在存在不确定性的情况下精化生物知识和机器人策略。其次，以新的方式利用估计和控制理论的工具(例如Cramer-Rao界)来分析鱼类跟踪算法的性能极限，并指导最优或接近最优的折衷设计以满足多个竞争目标，同时适应船上资源约束。在生物学方面，利用机器人网络对标记的鱼类进行连续、动态的跟踪是声学遥测技术向前迈出的重要一步，并为推进鱼类运动生态产生了新的数据集和模型。

Tracking Fish Movement with a School of Gliding Robotic Fish This project is focused on developing the technology for continuously tracking the movement of live fish implanted with acoustic tags, using a network of relatively inexpensive underwater robots called gliding robotic fish. The research addresses two fundamental challenges in the system design: (1) accommodating significant uncertainties due to environmental disturbances, communication delays, and apparent randomness in fish movement, and (2) balancing competing objectives (for example, accurate tracking versus long lifetime for the robotic network) while meeting multiple constraints on onboard computing, communication, and power resources. Fish movement data provide insight into choice of habitats, migratory routes, and spawning behavior. By advancing the state of the art in fish tracking technology, this project enables better-informed decisions for fishery management and conservation, including control of invasive species, restoration of native species, and stock assessment for high-valued species, and ultimately contributes to the sustainability of fisheries and aquatic ecosystems. By advancing the coordination and control of gliding robotic fish networks and enabling their operation in challenging environments such as the Great Lakes, the project also facilitates the practical adoption of these robotic systems for a myriad of other applications in environmental monitoring, port surveillance, and underwater structure inspection. The project enhances several graduate courses at Michigan State University, and provides unique interdisciplinary training opportunities for students including those from underrepresented groups. Outreach activities, including robotic fish demos, museum exhibits, teacher training, and "Follow That Fish" smartphone App, are specifically designed to pique the interest of pre-college students in science and engineering. The goal of this project is to create an integrative framework for the design of coupled robotic and biological systems that accommodates system uncertainties and competing objectives in a rigorous and holistic manner. This goal is realized through the pursuit of five tightly coupled research objectives associated with the application of tracking and modeling fish movement: (1) developing new robotic platforms to enable underwater communication and acoustic tag detection, (2) developing robust algorithms with analytical performance assurance to localize tagged fish based on time-of-arrival differences among multiple robots, (3) designing hidden Markov models and online model adaptation algorithms to capture fish movement effectively and efficiently, (4) exploring a two-tier decision architecture for the robots to accomplish fish tracking, which incorporates model-predictions of fish movement, energy consumption, and mobility constraints, and (5) experimentally evaluating the design framework, first in an inland lake for localizing or tracking stationary and moving tags, and then in Thunder Bay, Lake Huron, for tracking and modeling the movement of lake trout during spawning. This project offers fundamental insight into the design of robust robotic-physical-biological systems that addresses the challenges of system uncertainties and competing objectives. First, a feedback paradigm is presented for tight interactions between the robotic and biological components, to facilitate the refinement of biological knowledge and robotic strategies in the presence of uncertainties. Second, tools from estimation and control theory (e.g., Cramer-Rao bounds) are exploited in novel ways to analyze the performance limits of fish tracking algorithms, and to guide the design of optimal or near-optimal tradeoffs to meet multiple competing objectives while accommodating onboard resource constraints. On the biology side, continuous, dynamic tracking of tagged fish with robotic networks represents a significant step forward in acoustic telemetry, and results in novel datasets and models for advancing fish movement ecology.