EMG control of Robotic Arm

[Andy55]

My group and I are currently working on our senior design project that involves using EMG signals to control servo motors on a robotic arm.  We are using a BioRadio to record EMG signals from the biceps and forearm extensor muscles. We are able to connect the BioRadio to Labview to input the EMG data.  We have also been able to create a instrument driver for our robotic arm that allows us to control the servo motors in LabVIEW.  Yet, we are unable to figure out how to translate the EMG signals into actual robotic arm movement. Does anyone have any ideas on how to translate raw EMG data into values that can be used to control the robotic arm?

[Jazlan]

Hello Andy55

Perhaps this article by Hidalgo may be useful. http://bibdigital.epn.edu.ec/bitstream/15000/9302/2/P0180.pdf

I did a a project to control an RC car wirelessly using EMG signals, instead of the BioRadio, i used a circuit which consists of an instrumentation amplifier, band pass filter (20-500Hz), precision half wave rectifier and finally i used a smoothing low pass filter with very low cut off frequency (0.48). The output of this circuit was sent to a microcontroller. I used a program based on thresholding. I measured EMG from 2 muscles, when the amplitude of the EMG signal from the 1st muscle exceeded a threshold value, I programmed the DC motors of the RC car to move clockwise. When the amplitude of the EMG signal from the 2nd muscle exceeds a threshold value, the motors move counterclockwise.

Regards

Jazlan

[beirut]

My group and i are working on the same issue.

In order to classify different motions you need to use a classifier, we are using neural network in to order to make the decision on the type of the motion.

The neural network takes values extracted from the raw EMG signals collected from the muscles as an input, and gives you the decision.

The feature we are extracting from the EMG signals are:

1- Mean absolute value.

2-number of zero crossing.

3-number slope sign changes.

4-wave length of the signal.

5-cofficients of auto regressive model for the signal.

Hope this helps

[Steve.J]

A simple way to translate EMG into a workable signal is to rectify (absolute value in software) and low-pass filter with very low cut-off (as Jazlan used above). Alternatively, you can just compute the RMS power value of the signal.  In either case, you get a smoothly varying signal that corresponds to the general "power" of the EMG signal which is proportional to the intensity of the muscle recruitment.  What you do with this control signal is then up to you - you could interpret this control signal as a position signal (no effort equals baseline position, increasing effort tracks to some new position).  Another option would be to have the EMG intensity map to a force value for your actuator, or even a velocity control.  All depends on what your control scheme is.  Additional muscles/EMG signals could add additional axes of control.

[engali]

Hello Beirut

Could you please tell me if you are using Labview to extract the 5 feature types that you mentioned as well as neural networks to classift the EMG signals?  If not, what you are using  what software you are uning to perform feature extraction and classifcation of myoelecric signals?

[beirut]

Hello engali

No am not using labview to extract these signals, i used NI elvis scope in order to see the signal, but am extracting these signals using dspic from microchip.

You can write your code either in assembly or in c language.

[ZJ.Gu]

Hi engali,

See this article - http://decibel.ni.com/content/docs/DOC-8206.

Thanks!

[sudheer2011]

Hello beirut,as u have mentioned in third post about classifiers,can u tell me how to design a classifier in order to classify different motions?Thanx

[sudheer2011]

Can anybody tell me how to design a classifier in labview to classify different motions in labview after feature extraction is done?Thanx

[shirshendu96]

Can I use DAQ instead of a bioradio? and even if I use pre-amp, filter and amp is there an alternative to connecting it to a microprocessor? and even if I use DAQ do i have to filter and amplify the signal from surface emg?