REFLEXES: A Co-Designed Architecture for In-Network Control

近年来，传感、通信、控制和计算等领域的发展催生了一类新兴的复杂应用，称为信息物理系统（CPS）。在CPS中，过程不再局限于特定的单个设备，而是在通过通信网络互连的多个相互依赖的节点之间进行协调。CPS正在进入越来越多的领域，如工业自动化、能源、运输、CPS的部署承诺了相当大的好处，包括增加的灵活性、效率和适应性。然而，传统的通信和控制架构的分离和分层设计阻止了CPS的普遍采用，由于通信引起的不可靠性和延迟损害了通信质量，控制，甚至可能导致危险的行为。本项目的目标是开发一种新的合作，我们设计了通信和控制架构，以在给定可用通信和计算资源的情况下实现CPS的最佳性能。我们的核心概念在自然界中有一个类比：在身体内，反射是位于脊髓中的快速、预定义、简单的动作模式，即，密切（在低延迟的意义上）到传感器和执行器。更高级别的规划和控制机制位于大脑中更远的地方。在我们项目的第一阶段，我们引入了网络控制的新范式，它反映了自然的反射，通过利用（有限的）主动网络组件的计算能力。我们已经表明，当网络可靠并且传感器读数和控制器信号之间的关系简单时，这显著提高了控制质量。对于阶段II，我们的目标是将我们的方法扩展到更现实的场景。我们专注于提高无线控制系统的可靠性和网络中的高级处理功能，我们将考虑在控制器设计中用于不确定系统的先进鲁棒控制机制，包括更复杂的信号处理，考虑到有限的计算能力的网络元素和延迟引起的无线通信。为了允许确定性的控制器执行，我们将利用空间合作，以维持低控制消息延迟时，无线信道受损。我们因此避免不必要的延迟和不可靠性内封闭的控制回路，REFLEXES的关键方法贡献是（i）一种新的方法来对给定的通信和计算基础设施进行分布式控制，包括用于性能评估的适当分析工具，以及（ii）通过在单一的和根本上新的方法中共同设计控制和通信过程，在实时约束下用于网络内处理的新颖软件框架。

Recent advances in sensing, communication, control, and computation have fostered an emerging class of complex applications, called Cyber-Physical Systems (CPS).Therein, processes are no longer bound to specific single devices but coordinated between several interdependent nodes interconnected via communication networks.CPS are making inroads into an increasing number of domains such as industrial automation, energy, transport, and healthcare.The deployment of CPS promises considerable benefits, including increased flexibility, efficiency, and adaptability.However, the traditional separate and layered design of communication and control architectures prevents the ubiquitous adoption of CPS, as communication-induced unreliability and latencies compromise the quality-of-control and may even lead to dangerous behavior.The goal of this project is to develop a novel co-designed architecture for communication and control to enable the best possible performance of CPS given the available communication and computation resources.Our core concept has an analogy in nature: Within the body, reflexes are fast, pre-defined, simple action patterns located in the spinal cord, i.e., close (in the sense of low latencies) to the sensors and actuators.Higher-level planning and control mechanisms are located farther away in the brain.In phase I of our project, we introduced the novel paradigm of in-network control, which mirrors natural reflexes by allowing to push control functionalities as close as possible to a process by leveraging the (limited) computational power of active network components.We have shown that this significantly improves the quality-of-control when networks are reliable and relations between sensor readings and controller signals are simple.For phase II, we aim at extending our approach to more realistic scenarios.We focus on increasing the reliability of wireless control systems and on advanced processing functionalities within the network, both fundamental to real-world deployments.We will consider advanced robust control mechanisms for uncertain systems in controller design, including more complex signal processing, taking into account the limited computational power of in-network elements and the delay induced by wireless communication.To allow deterministic controller execution, we will harness spatial cooperation to maintain low control message latencies also when wireless channels are impaired.We thus avoid unnecessary latencies and unreliability within closed control loops, yielding significantly better quality-of-control.The key methodical contributions of REFLEXES are (i) a novel approach to distribute control on a given communication and computation infrastructure including appropriate analysis tools for performance evaluation, and (ii) a novel software framework for in-network processing under real-time constraints by co-designing control and communication procedures in a single and fundamentally new methodology.