Network-Informed Control - Control-Informed Network: towards multi techNology dynamICally ChangIng networks(NICCI^2)

网络动力系统的反馈控制提供的机会是巨大的。然而，如何利用现代无线通信、网络和计算技术来开发高性能网络物理系统尚不清楚。主要的障碍源于对组成部分的理解与将它们组合在一起时面临的挑战之间存在的巨大差距。关键的复杂性之一是控制环路的流量需求通常独立于通信能力来考虑。在当前的设置中，控制系统必须简单地应对网络提供的服务质量。虽然这有利于通信和控制系统的独立设计，但可实现的性能是有限的。因此，目前大多数缓慢且非安全关键的任务都是通过无线通信执行的。这些限制可以通过控制和通信系统的有效集成整体优化来克服。不幸的是，由于整个系统的复杂性和分布式特性，该优化问题的实际解决方案是不可行的。相反，我们努力寻求模块化解决方案。控制器和网络资源管理是分开设计的，但可以交换信息并协作交互。这种交换包括对控制要求、通信能力和技术选择的预测。在控制和通信资源管理方面开发的算法利用预测并从各自的行为中学习。在该项目的第一个融资阶段，我们分析了此类网络信息物理系统可实现的性能的限制。使用适当的确定性和随机框架，我们能够对网络建立最低要求，这是实现所需控制质量所必需的。我们进一步提出了许多控制感知调度和网络感知控制方法，其中网络管理器和控制器之间的信息是偶尔交换的。我们的见解激发了在第二个资助期内调查许多有趣且相关的研究问题。在该项目的第二个资助阶段，我们的目标是进一步阐明基本权衡并开发用于设计网络网络物理系统的模块化方法。我们将重点关注具有时间和状态相关属性的异构和互补无线通信技术可用的实际案例。将这些技术集成到可能具​​有反馈的大规模网络物理系统中是非常重要的。为了取得成功，我们将把信道预测、机器学习和估计技术与基于优化的调度和控制方法融合起来。我们将在联网车辆系统和机器人工业 4.0 测试场景上示范性地测试和完善我们的算法。

The opportunities provided by feedback control of networked dynamical systems are enormous. However, it is by no means clear how to harness modern wireless communication, network and computation technologies to develop high-performance cyber-physical systems. The main stumbling blocks stem from the significant gaps which exist between understanding of constituent parts and the challenges faced when bringing them together. One of the key complications is that the traffic demands of the control loops is often considered independently from the communication capabilities. In current setups, the control system has to simply cope with the quality of service provided by the network. Whilst this facilitates an independent design of communication and control systems, the achievable performance is limited. Hence, mostly slow and non safety-critical tasks are currently performed via wireless communication.These limitations could be overcome by an efficiently integrated overall optimization of control and com- munication systems. Unfortunately, a practical solution to this optimization problem is infeasible due to the complexity and distributed nature of the overall system. We strive instead for a modular solution. Controllers and network resource management are designed separately, yet exchange information and interact collabora- tively. This exchange comprises predictions of control requirements, communication capabilities and techno- logical choices. The developed algorithms at both the control and communication resource management sides, leverage predictions and learn from their respective actions.Within the first funding phase of this project, we analyzed limits of achievable performance of this class of networked cyber-physical systems. Using suitable deterministic and also stochastic frameworks, we were able to establish minimum requirements on the network, which are necessary to achieve the desired quality of control. We further proposed a number of control-aware scheduling and network-aware control methods wherein information between network manager and controller is exchanged sporadically. Our insights motivate numerous interesting and relevant research questions to be investigated within a second funding period.In the second funding phase of the project, we aim to further elucidate fundamental tradeoffs and develop modular approaches for the design of networked cyber-physical systems. We will focus on practical cases where heterogeneous and complementary wireless communication technologies with time and state-dependent properties are available. Integrating such technologies into possibly large-scale cyber-physical systems with feedback is highly non-trivial. To be successful, we will fuse channel predictions, machine learning and esti- mation techniques with optimization-based methods for scheduling and control. We will exemplarily test and refine our algorithms on networked vehicle systems and industry 4.0 test scenarios with robots.