Guiding Chaotic Swarm Dynamics in Evolving Networks of Agents with Privacy and Fairness Considerations

蜂群系统由许多独立的主体（机器人、人类、昆虫）组成，它们以协调的方式行动，以实现共同的目标。机器人群现在模仿群居昆虫或群居鸟类的行为，而人工智能代理在社交网络中相互作用。随着商用无人机和自动驾驶汽车等多智能体网络系统的不断兴起，物理和网络世界中的群体行为将变得司空见惯。因此，越来越需要开发方法来理解和控制群体行为的动态。智能体之间的相互作用主导着群体动力学，因此了解网络连通性的影响以及它是否会导致或控制群体中的混乱是至关重要的。此外，一些代理更倾向于维护信息隐私或价值公平，这将影响群体控制协议。这个项目关注的是开发和研究用于控制群体行为的引导网络进化博弈的动力学。应用包括在医疗应用中控制微型机器人群，以及管理在线社会现象以减少负面行为。该项目为年轻研究人员提供了宝贵的跨学科培训，并为高中生和本科生提供了有针对性的推广活动。本项目研究了由于网络演化动力学而产生的复杂混沌行为，开发了动态控制技术，并研究了这些系统中代理之间的隐私和公平相关问题。这项工作的重点是在网络中相互作用的半自主代理，并通过进化动力学改变它们的瞬时混合策略，重点是羊群和共识动力学。微观的、基于主体的行为与宏观的、进化方程（如网络复制因子和其他替代品）之间将有系统地建立联系。种群控制是通过控制基于网络的相互作用的博弈矩阵的周期性驱动来完成的。考虑代理隐私的控制系统动力学也将被考虑。虽然在性质上与经典控制中的不可观察性相似，但代理隐私是不同的，因为隐私影响代理之间的动态演化，而不是高级控制输入。在各代理机构之间公平分配或贡献资源方面也将采取类似的办法。该项目的目标是建立必要的控制理论初步，以允许平衡形成，以控制网络中相互作用的代理群体，确定这种设置下控制的理论极限，并了解代理隐私对系统进化的影响。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Swarming systems are composed of many independent agents (robots, humans, insects) that act in a coordinated fashion to accomplish common goals. Robotic swarms now mimic the behavior of social insects or flocking birds, while artificially intelligent agents interact in social networks. With the continuing rise of multi-agent networked systems, such as commercial drones and autonomous vehicles, swarming behaviors in the physical and cyber worlds will become commonplace. Consequently, there is an increasing need to develop methods for understanding and controlling the dynamics of swarming behavior. Agent interactions dominate swarm dynamics, so it is critical to understand the effects of network connectedness and whether it can cause or control chaos in the swarm. Additionally, some agents will prefer to maintain information privacy, or value fairness, which will affect swarm control protocols. This project is concerned with developing and studying the dynamics of guided networked evolutionary games for controlling swarming behavior. Applications include controlling microscopic robot swarms in medical applications and managing online social phenomena to decrease negative behaviors. The project provides valuable interdisciplinary training for young researchers, and targeted outreach activities towards high school students and undergraduates.This project investigates the complex chaotic behaviors that can emerge as a result of evolutionary dynamics on networks, develops techniques for dynamic control, and studies the problems associated with privacy and fairness among agents in these systems. The work focuses on semi-autonomous agents that interact with each other in a network and alter their instantaneous mixed strategies through evolutionary dynamics, with an emphasis on flocking and consensus dynamics. Connections will be made systematically between microscopic, agent-based behaviors, and macroscopic, evolutionary equations, such as the network replicator and other alternatives. Control of the population is accomplished by periodic actuation of the game matrix governing the network-based interactions. The dynamics of control systems that take agent privacy into consideration will also be considered. While qualitatively similar to non-observability in classical control, agent privacy is distinct because privacy affects the dynamic evolution among agents rather than the high-level control inputs. A similar approach will be taken to the fair allocation or contribution of resources among agents. The goals of this project are to establish the control-theoretic preliminaries necessary to allow equilibrium shaping in order to control populations of agents interacting in a network, to determine the theoretical limits of control in this setting, and to understand the impact agent privacy has on system evolution.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.