Using the robots is a big step forward. Just because you can do something in a computer doesn't mean you can do it the same way in real life. The networking problems and dealing with mechanical errors are new issues that will need to be tackled when we start dealing with even bigger large-scale physical bot swarms... hopefully ones with more realizable use cases.
Well put--there's a world of difference between yet another cutesy software sim of robots (of which I've been guilty!) and an actual hardware implementation. If you haven't actually dealt with things like shitty odometry, derpy wifi signals, packet corruption, motor driver noise screwing up logic, or any of a dozen other things, you probably don't appreciate just how tricky this is to get right.
If you'd like to see more about a lab doing real swarm stuff on real (buggy) hardware, check out the Multi-Robot Systems Lab at Rice:
Could you achieve some level of realism by introducing errors randomly at various input/outputs? Likely it wouldn't be as good as a hardware PoC, but I would expect it to be good enough.
Is this something used in simulations? I would think so.
So, I've been meaning to write a new sim for in-browser swarm manipulation and programming, taking into account things like message loss and noise in odometry readings, right? Seems quite reasonable on the face of it.
The problem is that you get genuinely weird stuff like IR comms just not working inside a torus of say 18 inches, but farther or closer working fine. Or issues where, when the motors both kick in at the same time, maybe a pin gets held a little lower for a little longer than it needs to be, and that causes an unrelated glitch (for a better story, look at http://www.quora.com/Software-Engineering/Whats-the-hardest-... ).
For any neat sim you make (and there are many!), people will inevitably just say, "Well that's all well and good, but have you tried it with robots?" Failing to do so consistently leads to a lab culture where you simply don't have the expertise in-house to do meaningful debugging, design, or research, because everybody's busy playing with their toy sims.
You'd think that that code would be still useful, but given the general quality of academic code and policies about reproducibility (which is to say, lol), pure sim work just disappears as a waste of money.
