Tutorial Session Proposal:
Communication Challenges in Networked Control Systems

Participants:

Dr. Girish Nair, University of Melbourne, Australia
Dr. James Moyne and Prof. Dawn Tilbury, University of Michigan, US
Prof. Karl H. Johansson, Royal Institute of Technology (KTH), Sweden
Prof. John Baillieul, Boston University, US

Rationale

For much of their history, communications and control have been regarded as largely separate disciplines. Though the advent of digital control led to the study of phenomena such as quantization noise, feedback policies were largely designed separately from the communication protocols that supported them. With high communication capacity this made sense, since overall analysis and design were simplified with negligible loss of control performance. However, as communications technology has become more ubiquitous, this modular approach has been challenged, first by the emergence of industrial control networks in the 1980s, and more recently by a growing number of applications in sensor networks and unmanned aerial vehicle coordination.

In these applications, multiple systems are controlled by using many geographically separated sensors and actuators communicating over a network of point-to-point links or a shared medium such as a wireless channel. The large number of transmitting nodes implies that even if overall communication capacity is large or latency is low, each node may effectively have only a small fraction of the available capacity, or have to compete with many other nodes for medium access. This can have a severe effect on the performance of the control policy, since quantization errors, varying communication delays, and bit errors become significant, and can then no longer be treated as small perturbations. Indeed, a major discovery of recent times was the existence of a minimum channel bit rate below which an unstable linear plant cannot be stabilized by any control policy.

Clearly, the full potential of a networked control system can only be achieved if the communication and control policies are not considered in isolation. The medium access policies, coding laws, and bit rate allocations should be designed with an eye on the overall control objective. Furthermore, as complete point-to-point connectivity between nodes is usually costly or impractical, it is crucial to understand the effect of network topology, and to characterize how information should flow through the system in order to achieve the desired goal, that is, which actuators need information from which sensors, and how quickly.

The aim of this tutorial session is to discuss some of the key theoretical and practical challenges in networked control. The session will commence with a 40 min talk (Nair) giving an overview of the basic theory of networked control with limited data rates, and also discussing some recent advances for noisy channels and cooperative networked control. This will be followed by three shorter talks (Moyne/Tilbury, Johansson and Baillieul) that will discuss emerging applications in
industrial control, automotive control, and mobile agent coordination. Abstracts of each talk are given below.

1. Networked Control with Limited Data Rates: an Overview
   Speaker: Girish N. Nair
   As networked control becomes more widely adopted, it is increasingly important to gain a clear understanding of the theoretical foundations of the field. This talk presents an overview of some of the key results on control systems with finite bit rates, channel errors and delays. The primary emphasis will on explaining how the interaction between the control- and information-theoretic concepts yields useful fundamental bounds and optimality conditions that can serve to guide practitioners. These include the Data Rate Theorem and its noisy channel extensions, the certaintyequivalence and quasi-separation principles for coded control, and various universal bounds that characterize how performance must vary with bit rate and delay. The importance of these results lies in their wide applicability, even when the feedback mechanisms are unclear, for instance in biological systems. The merits of specific schemes based on zooming quantization will also be discussed, in the context of noisy plants and channels. Finally, a very recent result on information flows in systems with multiple sensors and actuators will also be presented, together with its implications for network topology and bandwidth allocation.
2. Performance Considerations in the Application of Industrial Ethernet on the Factory Floor
   Speakers: James Moyne and Dawn Tilbury
   The most pervasive trend in networked control systems over the past three years has been the move towards Industrial Ethernet (IE) network solutions. IE is envisioned as a low cost, onesize-fits-all solution that allows capabilities such as internet services to be realized all the way down to the sensor level, while at the same time enabling factory-wide connectivity and resource coordination. While IE has many attractive capabilities, there are many performance issues that should be considered when evaluating the many facets of IE application and deployment. This presentation explores the current trends and issues associated with the use of IE and related network technologies for distributed, multi-level control, diagnostics and safety. This includes an exploration of various IE protocol technologies, a comparison of shared versus dedicated media approaches for network distribution of safety and control functionality, and a discussion of determining the cost associated with IE deployment and trade-off decision making. Results show that IE may not always be the best choice and the choice of IE over another network technology or an optimal IE configuration is often
   dependent on the application environment. The paper also presents ideas for future capabilities envisioned for IE including ideas for leveraging diagnostics networked systems to improve safety capabilities, as well as the exploring the move to wireless. Where appropriate, avenues for further exploration of these issues are identified, and the application of the concepts to a reconfigurable factory testbed (RFT) at the University of Michigan is explored.
3. Automation and Vehicular Applications of Networked Control
   Speaker: Karl H. Johansson
   An important feature of networked control systems is that they have to deal with limited and varying communication resources. These constraints depend heavily on the implementation technology. For example, in the control of transmission power for radio devices the bit rate needs often to be kept at a minimum restricting the obtained control performance, while in automotive control applications with data packets sent over a controller area network the difficulty is in predicting the communication demand for the variety of applications closed over the network. In this talk, we will
   motivate some recently considered research problems in networked control through a few specific applications in wireless automation, heavy duty vehicles and disaster relief.
4. Parsimonious Patterns of Relative Distance Sensing for Controlled Motions of Formations
   Speaker: John Baillieul
   In controlling the deployment of formations of mobile agents, it is frequently desirable to have each agent occupy and maintain a prescribed position relative to the other agents in the formation. (We draw a sharp distinction between the creation and movement of formations and flocking, the latter having rather loose relative distance relationships among the members of the flock.) Because in many applications the bandwidth available for inter-agent sensing and communications is limited, it is of interest to consider formation control strategies in which prescribing the relative
   position of each agent and carrying out the relative distance sensing needed to move the agents into and maintain their positions requires the minimum possible amount of communication and data processing. We shall discuss decentralized relative distance sensing control laws that are developed using recent new results in the theory of directed graph rigidity. Based on our own work over the past several years as well as recent work by Brian Anderson, Julien Hendrickx, Vincent Blondel, and others, we shall discuss the complete combinatorial classification of all structured planar formations. We shall indicate how this classification may be useful in addressing such problems as decentralized
   selection of a formation leader.