You might think that this is just another attempt to squeeze a video of a group of dancing Nao robots into the pages of an otherwise completely serious magazine. You might be right, but this isn't just another dancing robots news item - it's about a swarm of humanoid robots and how they interact. It's also fun to watch...
Sometimes I think that swarm roboticists just do it so that they have an excuse for getting lots and lots of hardware together in one place. When it comes to humanoid swarm robotics, the lure of being able to convince Aldebaran Robotics to lend eight Nao robots is quite understandable!
No it's not a Nao dance troupe - it really is a swarm.
It is collaboration between MIT's Nonlinear Systems Laboratory and Aldebaran Robotics. The scientific paper explains how a set of distributed oscillators can be brought into synchronism without using a central master oscillator. Each robot has its own oscillator which is synchronized to the robots around it using quorum sensing. A single robot is elected as the server and all of the other robots send it their current position in the cycle - their phase. It then computes the mean phase of the group and transmits this to the other robots. The other robots then adjust their oscillators to be closer to the mean. Slowly they converge to the same phase and synchronize.
This method of doing the job means that if a robot falls over or is taken out of the dance then, when it is put back, in it quickly re-synchronizes.
Now, or should that be Nao, watch the video:
It would be interesting to see what happens if the synchronizing method was changed from quorum sensing to the Kuramoto model - i.e. how fireflies come into sync. In this case there would be no master server gathering the votes of the others.
The whole idea that humanoid robots are cheap enough to consider the dynamics of swarms of them is something of a shock. Personally I still can't watch a swarm of Kilobots without feeling slightly creepy. A swarm of Nao's, so cute when they come in ones, could be equally creepy. I think the trick is to keep them dancing.
If you are want to apply data science techniques using parallel programming, in Apache Spark, you'll be interested in an edX course starting Monday June 1st that prepares you for the Spark Certif [ ... ]