Time-sensitive Control and Networking in Cyber-physical Systems

Organisers

  • John S. Baras
  • Karl H. Johansson
  • Mohammad Hossein Mamduhi

Speakers

  • Carlos Canudas-de-Wit, CNRS GIPSA-Lab, Grenoble, France
  • Konstantinos Gatsis, University of Oxford, UK
  • Maurice Heemels, Eindhoven University of Technology, The Netherlands
  • Alf Isaksson, ABB Future Labs, Sweden
  • P. R. Kumar, Texas A&M University, USA
  • Chrysa Papagianni, Nokia Bell Labs, Belgium
  • Adam Wolisz, Technical University of Berlin, Germany
  • Melanie Zeilinger, ETH Zürich, Switzerland

Summary

State-of-the-art application of cyber-physical systems (CPS) such as autonomous cars, health-care, robotics, and smart production, is characterized by time-sensitive functionalities, most notably, sensing, sampling, communication, control and actuation. This urges the development of real-time decision making methodologies such that the per-time requirements and limitations of both cyber and physical (application) layers are taken into account. In fact, a paradigm shift is essential from the traditional convention of designing stationary-like policies with asymptotic averaged performance guarantees to real-time synthesis of strategies that are capable of satisfying time-sensitive constraints across a heterogeneous network of users with time-varying objectives. To achieve this a tight integration of all communication, control and system components is needed to fully exploit their individual design elasticity and mutual adjustment potential. Recent evolution of communication technologies, e.g. 5G, has provided a great potential to achieve this goal by facilitating a fast, flexible, reliable, high-rate, and low-fault data exchange medium to support such coupled Net-CPS. The challenge lies in joint analysis and design of communication, control and systems methodologies that are aware of such network opportunities and exploit them to support time-sensitive application objectives. Motivated by this emerging interest, the proposed workshop will discuss the integration of control and communication in such constrained Net-CPS. The full day workshop will consist of three main topics, and a discussion panel. The main topics of interest are entitled as:

  • Time-sensitive control architectures for network-aware CPS
  • Time-sensitive networking architectures for control-aware CPS, and
  • New trends of industrial, commercial, and societal applications of CPS.

To cover the mentioned topics, talks will present recent works and future challenges in a broadly accessible manner. The workshop is concluded with panel discussion with the panel formed by academic and industrial experts. Our invited speakers include high profile senior and junior researchers with varied backgrounds in control and communication and computer science.

Programme

Scales Paradigms in Control & Estimation of Large-scale networks, Carlos Canudas-de-Wit

In this talk we presents some results from the ERC Scale-FreeBAck on the problem of model reduction, estimation and control of aggregates of large-scale systems with a limited number of measures/inputs. Aggregates here are “aggregated” variables functions of the systems state-space variables such as mean values. Examples of such a class of systems are traffic networks, Brain neural networks, heating systems, among others. The basic idea is to devise an aggregated model of the original large-scale system using the scale-free (SF) metric, which indicate that the degree distributions of the associated graph follows an exponential decaying law.  Then, we discuss different partitioning algorithms leading to aggregated graphs with the SF desired distribution but also with the suited control/observation properties. In the talk, I also present the mathematical properties necessary for the average observability. The second part of the talk, presents a different alternative for cutting system complexity, which consist in representing a large traffic network as a continuum. That is, to approximate a large-scale dynamic graph (where each node represent a variable), by   Partial Differential Equations. The objective of this second approach is to use the PDE model for designing boundary estimators and control.


Control and Learning for the Internet-of-Things, Konstantinos Gatsis

Emerging autonomous applications in the Internet-of-Things are enabled by sensing, processing, learning, and actuation capabilities joined by wireless communication. These connected communities of data-driven agents are in close interaction with the physical world as they collect data signals from sensors, learn and adapt to disturbances, coordinate, and actuate in an autonomous closed-loop fashion. Application domains include future transportation systems, industrial automation, robotics, and smart cities. These applications give rise to both new fundamental challenges and opportunities where the control community research can have significant impact. In this talk I will discuss the challenge of rethinking control at the edge of the Internet-of-Things and how this gives rise to novel co-design interfaces, as well as how statistical learning techniques can help cope with uncertainty and lack of models.


Event-triggered Control and Communication in Net-CPS, Maurice Heemels

Computer and communication technologies are developing at tremendous speeds leading to an increasingly networked and wireless world. These technologies bring new opportunities for the control of (large-scale) interconnected systems, but also raise new challenging questions for the resulting networked cyber-physical systems (Net-CPS) in which more and more data is exchanged. One important question is how to determine the time instances at which data should be exchanged between (sub)systems, agents, sensors, controllers and/or actuators such that desirable closed-loop stability and performance properties are guaranteed, while keeping the number of transmissions small. To efficiently use the available communication resources, we will discuss the potential of moving from traditional time-triggered control and communication schemes in which transmissions take place periodically in time, to event-triggered schemes. Event-triggered schemes form alternative and (more) resource-aware paradigms, as it seems natural to trigger control and communication actions by well-designed events involving the system's state, output or any other locally available information. In this talk we will discuss the main ideas, recent insights, and open questions for event-triggered control and communication schemes in the context of net-CPS. A general perspective will be provided for event-triggered communication (ETC) with side trips to consistency of ETC (in the sense of formal guarantees of outperforming periodic time-triggered communication), to dealing with network artefacts (delays, packet losses) and security issues in ETC, to handling the presence of disturbance and measurement noise, and to experimental validation of ETC in robotic and cooperative driving applications.


Industrial Requirements on Time-sensitive Control and Networking, Alf Isaksson

Control systems for process and manufacturing industries are right now on the brink of the largest shift since the introduction of the distributed control system (DCS) in the 1970s. This is driven by a dramatic increase in computational capacity in low cost devices including AI chips, as well as advances in communication such as time sensitive networking (TSN) and 5G. Meanwhile the introduction of cloud computing means that many computations that are today carried out inside the factory may in the future be carried out in the cloud. Hence a future control system will be a trade-off between cloud and edge/fog computing. This talk will discuss implications of this paradigm shift, and provide examples of applications where this may have a profound future impact. It will also try to describe the particular challenges that would arise if the system is using open standards in a much more decentralized architecture than today. Such challenges include cyber security as well as how to engineer and configure the system. Already today there is an increased trend towards verifying the performance of the control system prior to commissioning (often referred to as virtual commissioning). Furthermore, triggered by the advances in AI and autonomous driving, the notion of autonomy is discussed also for manufacturing industry. We will therefore briefly discuss what influence this may have on the future cyber-physical systems.


To be confirmed, P. R. Kumar


5G and Industrial Automation, Chrysa Papagianni

5G systems are set out to address the business contexts of 2020 and beyond, by enabling new network and service capabilities, opening up innovation opportunities for vertical markets. Communication service providers should be able to provide tailor-cut solutions to service requests from the verticals over the same network infrastructure. Network slicing provides a solution towards this end; it is considered a key for meeting 5G’s diverse requirements, including future-proof scalability and flexibility. However, to support the multiplicity of applications and services, we need to automate and optimally orchestrate networks through dynamic network slicing. Automation should be enabled throughout all phases of the network slice lifecycle through optimized closed-loop control at different levels and time-scales. In this talk, we will discuss the envisioned service automation architecture and focus on aspects related to orchestration and data plane programmability. The aforementioned topics constitute the research agenda of the Nokia Bell Labs Access Network Control department in Antwerp, Belgium.


Towards Cooperative Usage of Wireless Spectrum, Adam Wolisz

Cyber physical systems using wireless connectivity are dependent on spectrum availability. A lot of attention is actually devoted to wireless systems operating in licensed spectrum and aiming to meet the CPS time/quality requirements using the programmable networks/ network slicing paradigm. powerful alternative is, however, offered bin the form of cooperative usage of wireless spectrum by Software Defined Radio (SDR) based heterogeneous wireless technologies. In this talk advantages of the cooperative approach will be summarized, followed by a discussion of challenges as well as existing/envisioned solutions for efficient spectrum sharing supporting also  time critical CPS


Towards Safe Learning in Distributed Control Systems, Melanie Zeilinger

Many automated systems are connected in their objective, dynamics, or constraints to other systems via a system network, making it beneficial for the controller design to leverage coordination between systems for enhancing overall system performance. This task is, however, often challenged by the resulting complexity of the networked system, as well as computation and communication limitations. Learning models and controllers from data and experience here provides an important opportunity to address this challenge but can often not maintain the desired properties of the system network. In this talk, I will discuss techniques that can augment learning-based controllers with safety properties in the form of constraint satisfaction for system networks. The methods avoid a centralized controller design by only utilizing local system information, while aiming to reduce conservatism.