Overview

This project provides an interface to communicate with the Thalmic Myo, providing the ability to scan for and connect to a nearby Myo, and giving access to data from the EMG sensors and the IMU. For Myo firmware v1.0 and up, access to the output of Thalmic's own gesture recognition is also available.

The code is primarily developed on Linux and has been tested on Windows and MacOS.

Thanks to Jeff Rowberg's example bglib implementations (https://github.com/jrowberg/bglib/), which helped me get started with understanding the protocol.

Requirements

• python >=2.6
• pySerial
• enum34 (for Python <3.4)
• pygame, for the example visualization and classifier program
• numpy, for the classifier program
• sklearn, for a more efficient classifier (and easy access to smarter classifiers)

Dongle device name

To use these programs, you might need to know the name of the device corresponding to the Myo dongle. The programs will attempt to detect it automatically, but if that doesn't work, here's how to find it out manually:

• Linux: Run the command ls /dev/ttyACM*. One of the names it prints (there will probably only be one) is the device. Try them each if there are multiple, or unplug the dongle and see which one disappears if you run the command again. If you get a permissions error, running sudo usermod -aG dialout $USER will probably fix it.

• Windows: Open Device Manager (run devmgmt.msc) and look under "Ports (COM & LPT)". Find a device whose name includes "Bluegiga". The name you need is in parentheses at the end of the line (it will be "COM" followed by a number).

• Mac: Same as Linux, replacing ttyACM with tty.usb.

Included files

myo_raw.py (access to EMG/IMU data)

myo_raw.py contains the MyoRaw class, which implements the communication protocol with a Myo. If run as a standalone script, it provides a graphical display of EMG readings as they come in. A command-line argument is interpreted as the device name for the dongle; no argument means to auto-detect. You can also press 1, 2, or 3 on the keyboard to make the Myo perform a short, medium, or long vibration.

To process the data yourself, you can call MyoRaw.add_emg_handler or MyoRaw.add_imu_handler; see the code for examples.

If your Myo has firmware v1.0 and up, it also performs Thalmic's gesture classification onboard, and returns that information. Use MyoRaw.add_arm_handler and MyoRaw.add_pose_handler. Note that you will need to perform the sync gesture after starting the program (the Myo will vibrate as normal when it is synced).

classify_myo.py (example pose classification and training program)

classify_myo.py contains a very basic pose classifier that uses the EMG readings. You have to train it yourself; make up your own poses and assign numbers (0-9) to them. As long as a number key is held down, the current EMG readings will be recorded as belonging to the pose of that number. Any time a new reading comes in, the program compares it against the stored values to determine which pose it looks most like. The screen displays the number of samples currently labeled as belonging to each pose, and a histogram displaying the classifications of the last 25 inputs. The most common classification among the last 25 is shown in green and should be taken as the program's best estimate of the current pose.

This method works fine as long as the Myo isn't moved, but, in my experience, it takes quite a large amount of training data to handle different positions well. Of course, the classifier could be made much, much smarter, but I haven't had the chance to tinker with it yet.

myo.py (Myo library with built-in classifier and pose event handlers)

After you've done training with classify_myo.py, the Myo class in this file can be used to notify a program each time a pose starts. If run as a standalone script, it will simply print out the pose number each time a new pose is detected. Use Myo.add_raw_pose_handler (rather than add_pose_handler) to be notified of poses from this class's classifier, rather than Thalmic's onboard processing.

Tips for classification:

• make sure to only press the number keys while the pose is being held, not while your hand is moving to or from the pose
• try moving your hand around a little in the pose while recording data to give the program a more flexible idea of what the pose is
• the rest pose needs to be trained as a pose in itself

Caveats/issues

• on Windows, the readings become more and more delayed as time goes on
• doesn't have access to Thalmic's pose recognition (for firmware < v1.0)
• may or may not work with a Myo that has never been plugged in and set up with Myo Connect
• classify_myo.py segfaults on exit under certain circumstances (probably related to Pygame version)