Compact emg tokens cut wireless data while keeping hand movement accuracy
EMG Tokenization Preserves Movement Information at a Fraction of the Bitrate
Human-Computer InteractionMachine Learning
Summary
Wireless sensors that read muscle signals (EMG) usually send a lot of data, limiting battery life and speed. The researchers found ways to shrink this data into small 'tokens' without losing the ability to figure out precise 3D hand movements. They tested two methods to create these tokens and showed they almost match full data accuracy but use about 40 times less transmission bandwidth. This approach works well on various people and devices, and it runs fast enough on small processors. It means wearable muscle sensors can send less data while still allowing accurate movement control.
What this means in practice
- •For wearable device engineers: Implement data compression in muscle sensors to reduce wireless transmission without losing hand movement decoding accuracy.
- •For robotics control teams: Use compact EMG tokens to control robotic hands precisely with less data transfer and faster processing on embedded systems.
Authors
Raul C. Sîmpetru, Amin Olamazadeh, Devon Rohlf, Vlad Cnejevici, Alessandro Del Vecchio
Abstract
Wireless electromyography (EMG) interfaces must balance electrode coverage against transmission bandwidth. Here, we show that compact EMG tokens substantially reduce payload while retaining information for three-dimensional hand-pose decoding. We evaluate two causal tokenizers fitted without pose labels: an encoder pretrained to reconstruct EMG spectra and one based on SPAR, an engineered spectral-spatial representation of neighbouring electrodes. In personalized, within-session experiments with 320-channel EMG from nine participants, we compare tokenized and continuous features using matched decoder architectures and training procedures. Averaged across decoders, tokenization changed mean error by +0.01 mm for SPAR (95% confidence interval: -0.12 to 0.14) and +0.45 mm for the learned representation (0.10 to 0.82). Both increases met a 1 mm non-inferiority criterion. Stored token arrays reduced payloads 41-fold relative to raw EMG and 36-fold relative to continuous SPAR features. On sampled windows of the public emg2pose benchmark's test partition, absolute differences in mean error between learned tokens and continuous features were below 0.05 mm across 176 participants and in the subset of 20 participants unseen during training. Electrode coverage, calibration, and unseen recordings affected SPAR decoding more than quantization did. In a computational benchmark, optimized SPAR generated tokens for 63 channels within 20 ms on a Cortex-M55 microcontroller. These findings support tokenization as a compact interface between wearable sensing and movement decoding, allowing decoders to be updated while the wearable encoder and transmitted representation remain fixed.