Remove this ad

Open-Source MyoElectric signal pattern recognition using Machine Learning

Rss     Subscribe     Share     Tweet    

0 Points


Nov 22 16 4:10 PM

Tags : : , , , , , , , ,

I need help, does anyone knows how to code machine learning to recognize patterns in Myoelectric signals, you could change the world with this!Over the course of the last few months, a group of innovators from around the world have joined my daughter (who's left arm was paralyzed over the summer) and I to build an open source robotic assistive arm. This is our story so far, come join us and make it your story too!

Find more info on our facebook page:
Quote    Reply   
Remove this ad
Remove this ad

#1 [url]

Nov 24 16 10:36 AM

What sort of Machine Learning Algorithm do you wish to use? There are many! A Genetic Algorithm, evolutionary learning, neural networks, LDA etc. I'm actually going to program my masters project in Matlab to attempt to solve the problem of multiple gesture recognition at once. With a hand, we never just most one finger - they work independently. This is what I want to solve.

Quote    Reply   

#3 [url]

Feb 28 17 3:01 PM

We are currently working on the next version of the assistive arm that includes building the software as open source. I documented all we have done here so that others can do the same, and also documented our current challenges. Here is the Wiki where I am posting everything:

Quote    Reply   
Remove this ad

#4 [url]

Mar 1 17 11:50 AM


It's a great project.

I am curious as to why you're going to a machine learning approach over a standard 2-channel EMG approach (using signal amplitude to determine movement speed/strength). Is your reasoning because the weak signals have been hard to pick up with only 2 EMG sites? Or because you desire more functionality?


Quote    Reply   

#5 [url]

Mar 1 17 2:05 PM

We tried to use the standard 2 channel approach, but could not get a good enough signal, and even if we could it would be difficult to get that placement just right all the time. With the multiple electrode approach we are able to pick up the weak signals, also, using pattern recognition we could instead have the device quickly calibrate each time we put it on, and not need to worry about getting the electrodes at exactly the same location each time. Also the subtle variations in the body also changes how the signal is picked up, and so a quick calibration is needed each time. I am by no means an expert, but this is what we have learnt from experts in the field.  

Quote    Reply   

#6 [url]

Mar 2 17 11:08 AM

OMac wrote:

Have you managed to complete your multiple gesture recognition project?


I haven't started working on it yet, but I've spoken to my lecturers and my original idea on how to do it seems like it should work. I'm currently sitting an Applied Computational Intelligence module, and I'm going to prototype a very basic version to see if the theory will work. It's based on Euler's theory where any signal can be decomposed into coefficients, and then composed back to their original form. Based on that theory, I'm going to record individual finger movements, get their coefficients - and then store them in memory. Then using testing data I'm going to do the same thing, and then use an approximation function to see how close they are. The interesting thing then is to see if I can decompose multiple coefficients at once! 

Quote    Reply   
Remove this ad
Add Reply

Quick Reply

bbcode help