Hello, actually I'm starting to experiment with it. I just bought week ago. By this time I couldn't connect it to Max Msp, I'm looking what is going wrong. However, on powerpoint presentations it seems good. As always, you have to manage it and study perfectly which movement you want to make... Maybe, it's a little bit more difficult than leap motion. But, here we go... Let's see what other people say.

After a bit of research, I have the impression that the EMG data, which is supposed to be the special thing about it (aside from that, it's just an accelerometer and a gyroscope, which your phone already has) isn't really that useful. It has only five factory-recognized hand gestures, and from what I've seen online, there aren't loads of other ones that are easy to teach it.

From myo's developer blog:

Working with EMG signals is hard. Building an application that works for you and your friends is pretty straightforward, but building something that will work for everyone is another question entirely. The signals that are produced when different people make similar gestures can be wildly different. Different amounts of hair, fatty tissue, and sweat can impact what the signals look like, and this is compounded by the fact that the Myo armband can be worn on either arm and in any orientation. It’s incredibly difficult to transform a person’s hand into a controller, and we need to make sure that the applications in the Myo Market yield great experiences for users.

I found someone online using ML to teach the myo additional hand gestures, and they successfully added three. So for eight hand gestures total, not sure if it's worth it. At any rate, I'll be interested in hearing about your experiences.....

\M

Atau Tanaka is using them pretty successfully (IMHO) in Max: https://www.youtube.com/watch?v=G6H1J2k--5I

Right. I note that he's using the raw EMG data basically to create pulse trains. From the description of the video:

First a direct audification of motor unit action potentials is heard as spikes. This stochastic pulse train reflects performer limb activity.

This is certainly cool, and that direct data sonification is kind of Atau's thing, but doesn't really tell me much about how yummy that EMG data would be for situations where you wanted predictable control out of it, which is my particular area of interest. Thanks for the link, though!

\M

Here's some more information: http://www.arj.no/2015/06/02/mumyo/

Machine learning is not a bad idea. In general, EMG is noisy because our muscles are noisy. Also, you really need to look at a group of EMG sensors. Using something like Wekinator should enable fairly reliable gesture recognition.

I've also just started playing with the Myo, and have been a bit disappointed with the usefulness of EMG data. I was hoping to detect smallish but repeatable finger movements, but that's definitely not doable. I've been playing around with Wekinator, and there might be some promise there... for what I wanted to do I think there's just not enough discernible signal to train with, but curious what other people come up with.

I created a little tool to simplify hooking up the Myo to Wekinator, in case anyone else here wants to take a stab at it.