Detection and Tracking of Multiple Dynamic Targets with Cooperating Networked Agents

在多智能体框架中，一组自主智能体在潜在的动态、危险甚至对抗性的环境中合作执行复杂的任务。一般来说，团队必须寻找并监控可能也在移动的目标，同时平衡监控任务和持续的勘探。这种设置被广泛地称为持续监控，通常出现在移动机器人应用程序和传感器网络中，但它出人意料地丰富，并涵盖了许多其他不太明显的领域。在这个项目中，我们将开发一组执行持续监测的自主代理的最佳行为或至少接近最佳行为的数学技术，并将该理论部署在跟踪多个生物大分子在活细胞内移动的背景下。除了具有广泛范围的基础数学研究外，该项目的目标是构建一种新的跟踪荧光显微镜，该显微镜将利用数学框架提供比现有仪器更好的速度、精度和吞吐量来跟踪单分子的动力学。本科生和研究生都将接受各种学科的培训，包括优化、控制理论、机器人学和显微镜。此外，该项目还包括通过开发为期一天的单分子成像模块，将其作为Nanocamp的一部分，扩展到波士顿大都市区的低收入第一代到大学生。Nanocamp是一个为期六周的住宿暑期计划，面向目标人口结构中的高二和大三学生。在动态、危险和可能是敌对的环境中控制和协调代理人是非常具有挑战性的，因为它涉及多个目标和相当数量的信息交流，通常具有严重的沟通限制。由于特殊控制策略的使用经常导致系统性能低下，本项目中提出的方法是使用优化方法来创建设计良好、合理的策略，这些策略可以保证令人满意的(如果不是最优的)行为。由于此类优化问题在计算上很难处理，而且它们的解决方案很少适用于在线可扩展的分布式实现，因此其具体目标之一是开发使用可在线优化的参数控制策略家族的近最佳、高效和不确定性稳健的方案。虽然项目的主要目标是一个严格的数学和广泛适用的框架，但它的开发将考虑到主要的激励应用，即跟踪多个单一生物大分子。在这种情况下，试剂是单独的共焦体积，每个都使用可编程阵列显微镜独立寻址和控制，目标是荧光标记的生物大分子。虽然分散实施通常是可取的，但单分子跟踪应用程序支持集中式解决方案，因为所有实施都在单个控制器上完成，因此该项目将重点放在集中式方法上。所开发的数学算法将在现场可编程门阵列(FPGA)器件上实现，并通过跟踪自由扩散的量子点进行实验测试。

In the multi-agent framework, a team of autonomous agents cooperates in carrying out complex tasks in an environment that is potentially dynamic, hazardous, and even adversarial. In general, the team must seek out and then monitor targets that may also be moving while balancing the monitoring task with continued exploration. This setting, broadly termed persistent monitoring, typically arises in mobile robotic applications and sensor networks, but it is surprisingly rich and encompasses a number of other, less obvious, domains. In this project we will develop mathematical techniques for the optimal, or at least near-optimal, behavior of a team of autonomous agents performing persistent monitoring and deploy the theory in the context of tracking multiple biological macromolecules moving inside living cells. In additional to foundational mathematical research with a broad scope, the project aims to construct a new tracking fluorescence microscope that will leverage the mathematical framework to provide significantly better speed, accuracy, and throughput than existing instruments for following the dynamics of single molecules. Both undergraduate and graduate students will be trained in a variety of disciplines, including optimization, control theory, robotics, and microscopy. In addition, the project involves outreach to low-income, first-generation-to-college students in the Boston metro area through the development of one-day modules in single molecule imaging that will be used as part of Nanocamp, a six-week residential summer program for rising high school sophomores and juniors in the target demographic.The control and coordination of agents in dynamic, hazardous, and possibly adversarial environments is highly challenging since it involves multiple objectives and a considerable amount of information exchange with often severe communication limitations. Since the use of ad hoc control policies frequently leads to poorly performing systems, the approach proposed in this project is the use of optimization methods to create well-designed, rational policies that can guarantee satisfactory, if not optimal, behavior. Because such optimization problems rapidly get computationally intractable and their solution is rarely amenable to on-line scalable, distributed implementations, one of the specific aims is to develop near-optimal, efficient, and uncertainty-robust schemes that use a parametric family of control policies that can be optimized on-line. While the primary project goal is a mathematically rigorous and broadly applicable framework, it will be developed with the primary motivating application in mind, namely tracking of multiple single biological macromolecules. In this setting, the agents are individual confocal volumes, each independently addressed and controlled using a programmable array microscope, and the targets are fluorescently-labeled biological macromolecules. While a decentralized implementation is in general desirable, the single molecule tracking application supports a centralized solution since all implementation is done on a single controller and thus the project will focus on the centralized approach. The mathematical algorithms developed will be implemented on field programmable gate array (FPGA) devices and tested through experiment by tracking freely diffusing quantum dots.