Abstract

Fine grained finger position tracking has many applications in healthcare, extended reality (XR), and sports science. MyoPose uses hobby electronics to read the muscle activation signals of the muscles in the forearm with surface electromyograpgy (sEMG) and predict the angle of the joints of the fingers using novel machine learning architectures. MyoPose includes two models: one based on an autoencoder, commonly used for vision tasks, and Myo-BERT, a model based on LIMU-BERT, a transformer-based models intended for use with inertial measurment unit (IMU) data. MyoPose and Myo-BERT are within \(7.5^\circ\) median accuracy other solutions with consumer grade hardware with a smaller amount of training data and sample window size.

Team

  • Andrew Fantino

Required Submissions