I doubt that you will find a rigorous definition, it is not only a moving target but also leaves the impression that people found that wireless sensor networks are not sufficiently hip anymore and a new buzzword is needed to sell their research.

My take is that there is no set of problems underlying what people now call CPS, you could as well just call it "systems." (Vehicular networks are CPS as well?)

So define you own security requirements, label it CPS, and no one will notice.

Thanks for your insight. Although that sounds like an attractive alternative, that's exactly what I'm trying to avoid - I don't want to define some arbitrary set of requirements and call it CPS. I guess the alternative would be to work on the individual "cases" (industrial control systems, smart grids) that I find interesting, and then put things together under CPS at a later point. That solution isn't as satisfying as working on the general problem, though.

As for your question- kind of. There are a lot of parallels at least, and I think it's arguable that vehicular networks can be included (especially when considering data-centric security). I don't think I'm the right person to say "yes, they are", though, as I'm asking this question looking for a definition of CPS.

There is no organization that defines what the term means, so your guess is as good as everyone else's. From what I saw vehicular networks can be easily seen as CPS, so technically you already work in the field. Because the topic is so broad (aka ill-defined), there is no single set of requirements, so choosing a single case/application makes much more sense.

The reason for CPS being so popular is that the NSF has a dedicated funding project for it. Maybe you can extract a definition from the program or from the research they fund.

I found: http://www.nsf.gov/pubs/2011/nsf11516/nsf11516.pdf

(Sections I and II should have some more definite answers than research papers or surveys, but from what I saw they basically say "We all know CPS, but now we will make them better because they are so important.")

CPS can basically subsumes anything and everything which has mainly two components: 1. Cyber (Computation and Networking) 2. Physical (Interaction capability with Physical Medium). Considering this broad definition, one may include Robotics, VANET, Smart Grid, IoT, Pervasive Computing, Autonomous Vehicles, Autopilot Flights, Autonomous Medical Equipment, Intelligent Sensing Network etc. Thus, CPS is more about a complete system's science and technology as a whole rather than the simply an engineering paradigm.

I agree to your initial definition "cyber-physical systems refers to any type of system where networked systems interact with the real world" if you mean physical world by "real world". In this sense, at least CPS is clearly differentiated from "the cloud", which will never interact with the physical world itself. While the cloud term is pointing towards abstraction away from hardware (among other things), CPS focuses on interactions with hardware.

Personally, I would even count systems with wireless communication as CPS (e.g. GPS), as the physical layer has an impact on security. I feel the term traditionally comes from the Industry/Infrastructure community though - which of course includes all topics that Rajesh mentioned.

While I agree with most of your posts on principle, I have some trouble with this broad definition, in particular the inclusion of robotics. The problem is that robotics is normally considered one of the main embedded systems applications. To put it bluntly; all embedded systems are focussing on hardware interaction: does this make all of them cyber-physicial systems?

In addition, robotics lacks some of the characteristics (at least traditionally, it may be that advances in the field have changed this): high degree of connectivity and a network that is at least partially accessible to the attacker.

I think, Rens is getting confused between embedded system and CPS. Here to mention that not all embedded systems can be categorized as CPS unless it has some interaction capability with the physical environment and can interact with other similar embedded system (networking).

However, as far as Robotics is concerned, the modern era robotics, which is going into the paradigm of cooperative operation, is certainly a candidate for CPS. These group of robots are having capability of interaction with its physical environment as well as communicate with its peers to achieve some common objective.

Thanks for letting me know about the IEEE Network special issue, it somehow slipped by me.