CAREER: Modeling and Design of Composite Swarming Behaviors

该项目从单个主体的属性(包括噪声传感器和执行器以及随机行为)入手，逐步建立复杂群体行为的模型，并通过物理系统的系统实验来验证模型，将弥合简单原语分析与其组成成为复杂群体行为之间的差距。具体地说，这项研究将包括对包容、划分和重新配置集群基元的全面研究，然后展示如何将它们组合成复杂的集群行为，如模式识别、基于传感器的运动和自适应形状变化。这些原语和行为之所以被选中，是因为它们挑战了现有的建模方法，并导致了独立的贡献，以应对重大挑战的应用程序。受蜜蜂、白蚁、蚂蚁和多细胞生物体的健壮和可扩展操作的启发，这项拟议工作的资金将导致一种使用概率模型以原则性方式构建随机群体行为的方法。这些模型将基于单个代理的感知、驱动、通信和计算属性的概率行为，通过在从物理实验、运动学模型和随机模拟到宏观差分方程组的多个抽象层次上对系统进行建模。通过将模型建立在物理实验中，模型将不仅能够作为计算工具，而且能够作为单个代理的物理属性的设计工具。广泛影响：建模和设计集群系统的原则性方法将为设计和部署用于搜救、环境响应、精准农业和监视等领域的集群机器人系统奠定基础。与此同时，设计群体的能力也可能有助于揭示生物和化学系统的工作原理，如群居昆虫、多细胞系统和自组装，从而能够设计出具有任意功能的复杂多细胞系统。最后，这项研究产生的算法和系统适合于艺术装置，将广大公众吸引到集群系统的魅力中，并培养了对所有生物系统共同的分布式本质的理解。拟议的研究将与教育深度结合。在这一综合研究和教育计划中开发的具体课程包括：向四年级学生介绍计算思维的模块化机器人活动，面向代表不足群体的一年级学生的嵌入式系统速成课程，面向高年级学生的综合机器人学课程，以及关于“群体智能”和“自我组装”的跨学科研究生研讨会，该研讨会将积极吸引来自其他学科的学生，并鼓励他们将拟议的多层次建模方法应用于他们的研究。

This project, developing models for complex swarming behaviors by incrementally starting from the properties of an individual agent, which includes noisy sensors and actuators and stochastic behavior, and validating the models by systematic experimentation with physical systems, will bridge the gap between analysis by simple primitives and their composition into complex swarming behaviors. Specifically, this research will consist of a comprehensive study of containment, partitioning, and re-configuration swarming primitives, and then show how they can be composed into complex swarming behaviors such as pattern recognition, sensor-based motion, and adaptive shape change. These primitives and behaviors have been chosen because they challenge existing modeling approaches and lead to independent contributions addressing grand challenge applications by themselves.Inspired by the robust and scalable operation of bees, termites, ants and multi-cellular organisms, the funding of this proposed work will lead to a methodology to compose stochastic swarming behaviors in a principled way using probabilistic models. These models will be grounded in the probabilistic behavior of an individual agent's sensing, actuation, communication and computational properties by modeling the system at multiple levels of abstraction going back and forth from physical experimentation, kinematic models, and stochastic simulations to macroscopic difference equations. By grounding the models in physical experiments, models will be able to serve as design tool for not only the computational, but also the physical properties of an individual agent.Broader Impacts: A principled methodology for modeling and designing swarming systems will create the foundation for designing and deploying swarm robotic systems for search and rescue, environmental response, precision agriculture, and surveillance, among others. At the same time, the ability to design swarms might also shed light on the workings of biological and chemical systems, such as social insects, multi-cellular systems, and self-assembly, enabling the design of complex multi-cellular systems with arbitrary functionality. Finally, algorithms and systems resulting from this research lend themselves to artistic installations, drawing a broad public into the fascination of swarming systems and nurture an understanding for the distributed nature that is common to all living systems. The proposed research will be deeply integrated with education. Specific offerings developed in this integrated research and education plan include: a modular robotic activity that introduces computational thinking to 4th graders, a crash-course on embedded systems geared at 1st year students from under-represented groups, a comprehensive robotics curriculum for upperclassmen, and an interdisciplinary graduate seminar on "Swarm Intelligence" and "Self-Assembly" that will actively involve students from other disciplines and encourage them to apply the proposed multi-level modeling methodology to their research.