Optimal Co-Design of Wireless Resource Management and Multi-Loop Networked Control

由计算算法控制或监视并由提供信息交换的通信网络支持的物理实体通常被称为网络物理系统(CP)。智能电网、自动驾驶和机器人技术是CPS最突出的一些例子。从系统论的观点来看，控制系统由多个控制回路组成，这些控制回路在共享的通信网络上闭合，也称为网络控制系统(NCS)。在典型的NCS场景中，控制层和通信层对彼此的性能有很大影响。需要新的灵活且可扩展的跨层解决方案，其响应各个层的实时变化。这种通信和控制策略的开发是我们研究项目的重点。在DFG SPP网络物理网络(CPN)项目的第一阶段，我们研究了共享无线单跳和两跳网络中的控制和通信协同设计问题，重点研究了多个线性二次型调节器(LQR)反馈控制环路和具有分组擦除信道模型的介质访问控制(MAC)层。我们开发了控制和通信策略，以使总体控制质量(QoC)最大化，并为NCS中控制和MAC策略的优化联合设计提出了一种初步的、根本新颖的系统解决方案。此外，我们与SPP CPN中的合作伙伴合作，开发了一个轻量级开源控制和通信平台，用于初始基准测试。在该项目的第二阶段，我们的目标是解决一个更一般的全局优化问题，该问题在考虑多跳网络中的转发延迟、排队和丢包以及控制任务的异构性的情况下，找到最优的控制和通信联合策略。作为我们方法的一部分，我们引入了信息效用(UOI)的新概念。我们的目标是开发将耦合效应、层间依赖关系和异类任务需求统一到一个功能中的UOI功能。UOI函数及其近似支持将全局问题分解为以分散方式解决的局部子问题。在我们的项目过程中，我们将为不同层的各种控制任务关键点和通信上下文建立对UOI的基本理解。我们开发了一个多跳多环控制网络的框架，通过信息交换对控制和网络状态以及相应的转换进行分析建模。最终目标是制定全局优化问题，使多循环和多跳场景的控制质量最大化，并推导出可处理和可扩展的本地策略。我们的目标是在实际的概念验证实验中验证我们的理论发现，多个机器人共享一个多跳无线网络。

Physical entities that are controlled or monitored by computational algorithms and supported by a communication network providing information exchange are typically called cyber-physical systems (CPS). Smart grids, autonomous driving and robotics are some of the most prominent examples to CPS. From a system theoretic viewpoint, CPS consist of multiple control loops closed over shared a communication network, also called networked control systems (NCS). In typical NCS scenarios, the control and communication layers strongly affect the performance of each other. Novel flexible and scalable cross-layer solutions are required, which are responsive to the real-time variations of the individual layers. The development of such communication and control strategies is in the focus of our research project. During the first phase of our project in the DFG SPP Cyber Physical Networking (CPN), we have investigated the control and communication co-design as an optimization problem in shared wireless single-hop and two-hop networks with a focus on multiple linear-quadratic regulator (LQR) feedback control loops and medium access control (MAC) layer with a packet erasure channel model. We have developed control and communication policies such that the overall quality-of-control (QoC) is maximized and presented an initial fundamentally novel systematic solution for the optimal co-design of control and MAC policies in NCS. Moreover, in cooperation with partners in the SPP CPN, we have developed a lightweight open source control and communication platform for initial benchmarking.In this second phase of the project, our goal is to solve a more general global optimization problem that finds optimal joint policies for control and communication taking forwarding delay, queuing, and packet loss in a multi-hop network as well as heterogeneous control task criticalities into account. As part of our approach we introduce the novel concept of utility-of-information (UoI). We aim at developing UoI functions that unify the coupling effects, inter-layer dependencies and heterogeneous task requirements in one single function. UoI functions and their approximations support the decomposition of the global problem into local sub-problems that are solved in a decentralized fashion. In the course of our project we will develop a fundamental understanding of the UoI for various control task criticalities and communication contexts of different layers. We develop a framework for multi-hop and multi-loop control networks that analytically models control and network states and corresponding transitions through information exchange. The final goal is to formulate the global optimization problem that maximizes the quality-of-control for multi-loop and multi-hop scenarios and to derive tractable and scalable local policies. We aim to validate our theoretical findings in a practical, proof-of-concept experiment with multiple robots sharing a multi-hop wireless network.