AitF: Collaborative Research: A Distributed and Stochastic Algorithmic Framework for Active Matter

蜂群机器人研究的是一群机器人如何朝着一个单一的目标工作。这样的目标通常是通过为每个机器人配备感官能力、基本计算能力和驱动来实现的。传感器检测到环境的一些信息，这些信息被用来决定下一个动作，并执行一些结果驱动。近年来，蜂群机器人技术取得了许多进步，但仍处于起步阶段。pi将采取“以任务为导向”的方法，从期望的宏观紧急集体行为开始，开发机器人将运行的分布式和随机算法基础（在微观层面），以便收敛到期望的宏观行为；作为这个过程的一部分，它们还将为尚未探索的集体和紧急系统提供理解。设想中的机器人规模很小，大小从毫米到厘米，因此当部署在拥挤（即密集）的环境中时，它们将表现为活性物质，更具体地说，是宏观的可编程活性物质。模拟中感兴趣的突发行为包括聚类（形成一个紧密连接的社区）、压缩（在最小化周长的同时保持连接社区的一致性）、群集（确定一致的方向）和运动（在保持凝聚力的同时集体移动）。其中许多都有有趣的反向问题，这些问题也同样值得考虑，例如探索（保持连接的种群，但探索最大面积）和分离（防止在二元粒子混合物中分离）。pi在跨学科研究方面有很强的记录，包括发起跨学科领域，例如，机器人物理学（Goldman），自组织粒子系统（Richa），以及统计物理学和随机算法的融合（Randall）。这些私人学院也致力于支持少数族裔、女性和本科生的研究（例如，通过亚利桑那州立大学的NSF S-STEM项目；佐治亚理工学院的ADVANCE和S.U.R.E.项目）。这个项目将汇集多个学科的技术，新的研究方法和发现将被纳入研究生课程。研究结果（包括开源代码）将在各个学科中发布，并将在web和ArXiv上提供。该项目的具体目标是努力开发一个面向任务的活动物质的理论框架，由简单物理系统模型提供信息，可以实现和测试算法。生物物理学家为理解自然而建立的群体机器人系统可以被修改，以执行这些新算法所要求的任务。物理模型将允许在额外的约束条件下对理论进行改进，例如重力和有限能量。它也将允许pi测试他们的算法的稳健性，因为物理系统承认一些错误。群体机器人的基本原理将从物理学的角度来研究，通过将集合视为微观层面上由可编程元素组成的活性物质。因此，将遵循（宏观）任务导向的方法，以设计一个分布式的随机算法框架，在微观层面构建和评估产生目标紧急宏观行为的算法。

Swarm robotics explores how groups of robots can work towards a singular goal. Such a goal is typically achieved by equipping each robot with sensory capabilities, basic computing power, and actuation. The sensors detect something about the environment, this information is used to make a decision about the next action, and some resulting actuation is performed. Swarm robotics has made many advances in recent years, but it is still in its infancy. The PIs will take a "task-oriented" approach and start from a desired macroscopic emergent collective behavior to develop the distributed and stochastic algorithmic underpinnings that the robots will run (at the microscopic level) in order to converge to the desired macroscopic behavior; as part of the process, they will also provide the understanding for yet unexplored collective and emergent systems. The robots envisioned are small in scale, ranging in size from millimeters to centimeters, so that when deployed in crowded (i.e., dense) environments, they will behave as active matter, more specifically as macroscopic programmable active matter. The emergent behaviors of interest for simulations include clustering (forming a tight-knit community that is mostly well-connected), compression (maintaining coherence of a connected community while minimizing perimeter), flocking (determining an agreed upon direction of orientation), and locomotion (collectively moving while maintaining cohesiveness). Many of these have interesting converse problems which are also equally worthwhile, such as exploration (maintaining a connected population, but exploring maximal area) and desegregation (preventing separation in a binary mixture of particles).The PIs have strong records for interdisciplinary research, including initiating interdisciplinary areas, e.g., robo-physics (Goldman), self-organizing particle systems (Richa), and the fusion of statistical physics and randomized algorithms (Randall). The PIs also have a strong commitment toward supporting minorities, women, and undergraduate research (e.g., through NSF S-STEM programs at ASU; ADVANCE and S.U.R.E. programs at Georgia Tech). This project will bring together techniques from multiple disciplines, and new research approaches and findings will be incorporated into graduate courses. Findings (including open source code) will be published in the various disciplines, and will be made available on the web and ArXiv.The specific goals of this project are to work toward developing a theoretical framework for task-oriented active matter, informed by models of simple physical systems, that can realize and test the algorithms. The swarm robotics systems that biophysicists build to understand nature can be modified to perform the tasks these new algorithms require. The physical models will allow refinements to the theories under additional constraints, such as gravity and limited energy. It also will allow the PIs to test their algorithms for robustness, as physical systems admit some error. The fundamentals of swarm robotics will be studied from a physics standpoint, by viewing the ensemble as active matter composed of programmable elements at the micro-level. Thus, a (macro-)task oriented approach will be followed in order to design a distributed, stochastic algorithmic framework to construct and evaluate algorithms at the micro-level that yield the targeted emergent macro-behavior.