My First Real-Time EMG Decoding of Over 20 Degrees of Freedom in the Human Hand

Surface electromyography (sEMG) is a non-invasive technique that measures electrical activity in muscles, making it valuable for prosthetics and assistive systems. However, current sEMG-based decoding algorithms are limited in the number of degrees of freedom they can simultaneously and proportionally control in real-time. This limitation restricts their utility in applications like advanced prosthetics and human-machine interfaces.

In this study, we present a deep learning method that decodes forearm muscle activity into proportional and simultaneous control of over 20 degrees of freedom of the human hand with real-time resolution and latency below 50 ms, matching neuromuscular delays. Using high-density sEMG data and hand kinematics recorded during various movements (grasping, pinching, digit-specific motions, and gestures) at different speeds, our neural network achieved real-time kinematic predictions at 32 updates per second. Employing transfer learning and a prediction smoothing algorithm, we reconstructed the full 3D geometry of the hand in real-time.

Our results demonstrate that high-density sEMG signals contain sufficient information to predict full hand kinematics, offering immediate potential for applications in individuals with motor impairments. This method represents a significant advance in decoding muscle activity for precise, real-time hand control.