Closing the Loop with IoT—the Role of Control

Tariq Samad
November 8, 2016


In this "first wave" of IoT, attention has concentrated on wireless sensors, cloud connectivity, big data analytics, and mobile apps. The concept of IoT, however, extends beyond these components and capabilities. For example, a white paper by the IEEE Internet of Things Initiative defines a much more expansive vision [1]:

"Internet of Things envisions a self-configuring, adaptive, complex network that interconnects 'things' to the Internet through the use of standard communication protocols. … The things offer services, with or without human intervention, through the exploitation of unique identification, data capture and communication, and actuation capability."1

This "actuation capability," especially "without human intervention," needs additional tools and research. Indeed, most definitions of IoT are from the perspective of information and communication technologies (ICT), but closed‐loop control in any context is not just, or primarily, an ICT challenge. Deep understanding of dynamics and control is essential. Feedback can qualitatively change the behavior of a dynamical system, for better or worse. A seemingly benign system can become unstable if feedback is inappropriately applied, and, on the other hand, automatic feedback control can enable unstable systems to reach levels of performance unattainable by stable systems. The closed‐loop integration of physical systems with the internet will require close collaboration between control experts and ICT experts.

Topics for control research

IoT also promises new vistas for the control research community. The fact that aircraft, cars, refineries, buildings, and medical devices function as well as they do is testament to the power and maturity of control science and engineering. But it's worth noting a few assumptions on which this success rests. The communication networks in control systems are generally assumed to be deterministic and reliable. Real‐time operating system platforms rely on predetermined, static schedules for computation and communication. Some control is now occurring over the internet, but at a supervisory level—for power grid distribution stations, wastewater treatment plants, some commercial buildings, and other applications. Closed-loop automation, more often than not, requires a dedicated, on‐site, end‐to‐end control system.

Control in the Internet of Things imposes control-theoretic challenges that we are unlikely to encounter in our usual application domains. More research is needed in a number of areas, including the following [2]:

  • Control over nondeterministic networks. Today's control systems assume deterministic communication and computation—in fact the execution and communication infrastructure is rigorously designed to ensure determinism. Nondeterminism—e.g., unpredictability in sensor reading, packet delivery, or processing time—complicates closed‐loop performance and stability.
  • Latency and jitter. Control over the internet and clouds will require much greater attention to latency (the end‐to‐end delay from sensor reading to actuation) and jitter (the variance in the intersampling interval). The techniques used in control applications today to deal with these phenomena are unlikely to suffice.
  • Bandwidth. Many control applications are not demanding of communication bandwidth—a few sensor reads and actuator outputs a second can suffice. But even this level of network performance may not be assured with mobile and/or internet connectivity. Furthermore, in the Internet of Things, closed‐loop control with feedback of video and other high-dimensional data is envisaged. The sophisticated signal and image processing algorithms involved will best be run on cloud platforms and will stress available bandwidth.
  • Cyber‐ and physical security, and resilience. The physics of the "things" in IoT, if appropriately incorporated, can enhance detection and protection approaches for both cyber and physical security. Conversely, physics and feedback can open the door to new attack scenarios: e.g., a well‐performing control system may be rendered unstable by introducing small delays in communication pathways.
  • Interoperable and plug‐and‐play sensors, models, and algorithms. With our digital devices and platforms we have become accustomed to features such as auto-discovery, search, composition of services, and plug-and-play integration. These are not as yet available for control applications. To get there, interoperability will need to extend beyond the interface specification; "dynamic" compatibilities will also be critical.

Related IoT enhancements

Today's IoT infrastructure places limitations that new theoretical and algorithmic developments by the control community can only partially overcome. Available component technologies and the IoT stack as often envisioned fundamentally limit the potential for advanced control. Research in IoT technologies is targeting these limitations and will also open the door for closed-loop control, especially for high-bandwidth, highly reliable, and high-performance applications:

  • 5G networks.Cellular communication technology has progressed by "generations" but the next advance is seen as a "paradigm shift" [3]. Dramatic enhancements in bandwidth, flexibility, and intelligence are foreseen, with data rates two to three orders of magnitude greater than 4G systems.
  • The tactile internet. Round-trip latencies with wireless communication are not currently low enough for many real-time control applications. A key threshold is seen as 1 ms, at which point human-in-the-loop wireless control becomes feasible. This era of the "tactile internet" is expected to open up a vast space of new closed-loop applications [4].
  • Fog computing. The cloud is a central element of today's IoT stack but a critical bottleneck for reliable real-time control. "Fog" or edge computing architectures enable processing to occur closer to the sensors and actuators, with advantages of speed, security, reliability, and efficiency [5].


Control expertise will be required to realize the visions of IoT that we, its proponents, are promising. At the same time, IoT brings new and exciting opportunities for research and development in control. Research in IoT platforms and technologies is also targeting enhancements that will provide the infrastructure necessary for supporting advanced real-time closed-loop applications.

To illustrate, here are some prospects that can motivate collaborative research [2]:

  • Systems that are not physically connected or collocated could be coordinated in real time;
  • Optimized performance (e.g., energy efficiency) could be achieved for small‐scale systems that cannot afford dedicated control systems;
  • High‐fidelity models could be widely applied for real‐time control via IoT implementations;
  • Global networks of sensors and actuators could be implemented and coupled with sophisticated control and optimization algorithms;
  • Greater redundancy and fault‐tolerance could be achieved across critical infrastructures.

There's much to be done before the full IoT vision can be realized. And control engineers and scientists have a critical role to play.

Parts of this article previously appeared as [6].



1 This excerpt is for the "large environment scenario" discussed in [1]; the relevance to closed-loop applications is also evident in the lower-complexity scenario that is also described.


[1 ]R. Minerva, A. Biru, D. Rotondi, Towards a definition of the Internet of Things (IoT), 27 May 2015 [accessed 22 Oct 2016]

[2] T. Samad, The Web of Things and Cyberphysical Systems: Closing the Loop, W3C Workshop on the Web of Things, Berlin, June 2014.  [accessed 22 Oct 2016]

[3] J.G. Andrews et al., What will 5G be?, IEEE Journal on Selected Areas in Communications, vol. 32, no. 6, pp. 1065-1082, June 2014

[4] G. P. Fettweis, The Tactile Internet: Applications and Challenges, IEEE Vehicular Technology Magazine, vol. 9, no. 1, pp. 64-70, March 2014. doi: 10.1109/MVT.2013.2295069

[5] Cisco, Fog Computing and the Internet of Things: Extend the Cloud to Where the Things Are, 2015.  [accessed 22 Oct 2016]

[6] T. Samad, Control Systems and the Internet of Things, IEEE Control Systems Magazine, vol. 36, no. 1, pp. 13-16, Feb. 2016. doi: 10.1109/MCS.2015.2495022



Tariq SamadTariq Samad holds the Honeywell/W.R. Sweat Chair at the Technological Leadership Institute, University of Minnesota. Until May 2016 he was with Honeywell, retiring as Corporate Fellow and Global Innovation Leader. His interests relate broadly to automation, intelligence, and autonomy for complex engineering systems. Dr. Samad was President of IEEE Control Systems Society and the American Automatic Control Council. He is a Fellow of IEEE and IFAC. Other recognitions include the IEEE CSS Control Systems Technology Award, an IEEE-CSS Distinguished Member Award, and an IEEE Third Millennium Medal. He is editor-in-chief of IEEE Press. Dr. Samad holds 20 patents and has over 100 conference, journal, and book publications. He has a B.S. from Yale University and M.S. and Ph.D. degrees from Carnegie Mellon University.




2016-11-08 @ 6:45 PM by Coote, Tim

Is a control model really new?  I had to include it in the architecture consumer facing IoT systems a couple of years ago.  An abstraction that may help is to consider the 'control systems' themselves as other 'Things' that form part of the IoT.  Such a model makes it more obvious that an aspect of IoT is the compute nodes, as well as sensors and actuators. It also makes it simpler to understand the need for peripatetic software.

There is already much work on how such distributed state can be provably consistent (e.g. embedded in the experimental systems based on E; and some work at HP-Labs: But there does need to be some research on how to fail-fast and fail-safe if visibility of relevant parts of the controlled system disappear.

Interestingly, some current consumer targeted controlling IoT systems have significant theoretical holes as they rely on 'eventual consistency' of dependency graphs.

Unfortunate system behaviour can also easily arise from dependencies that are not visible to the control system - or are too expensive to model accurately - , and, I concluded that a major issue is the simulation and testing of such systems at scale.

2016-12-13 @ 10:28 AM by Milis, George

Very interesting article and we fully second the need for control engineering to further exploit the IoT domain. We are currently working towards this direction and we have published some preliminary work in conferences, e.g.:

1. A Cognitive Fault Detection Design Architecture [WCCI2016, link to IEEE Xplore below](
2. A Cognitive Agent Architecture for Feedback Control Scheme Design [presented at SSCI2016 last week]

In few words, we combine background knowledge on logic systems and inference mechanisms and couple it with background logic in feedback control (also fault detection) systems so as to enable online configurations that consider all sub-systems as IoT services.