Continual learning improves cross-user muscle signal gesture recognition
Cross-User Adaptation for sEMG Gesture Recognition via Continual Learning
Machine Learning
Summary
Recognizing hand gestures from muscle signals (sEMG) is hard because everyone's muscles produce different signals. The authors studied new ways to keep computer models updated as they learn from different people and new gestures without starting over. They found one method, called experience replay, works best to remember old gestures while learning new ones. Also, how the data is fed into the system matters a lot for stable learning. Adding handcrafted features helps accuracy but only with experience replay to keep performance over time.
What this means in practice
- •For prosthetic device engineers: Update hand gesture recognition models with new users and gestures without retraining from scratch, improving adaptability in prosthetic control.$Commercial implications: Enables adaptive prosthetic controllers that maintain accuracy across users and gesture updates, making more reliable, personalized devices for amputees.
- •For wearable sensor developers: Implement continual learning with shuffled training to enhance gesture recognition stability and accuracy across diverse users.
Tested on one dataset.
Authors
Aran Ramakrishnan, Shubhi Mishra, Tyler Ham, Assel Kembay, Jason Eshraghian
Abstract
Prosthetic and assistive interfaces increasingly depend on decoding hand gestures from surface electromyography (sEMG). Because the signal is shaped by person-specific factors, the recurrent neural networks (RNNs) commonly used to model these temporal streams must adapt to new wearers and new gesture classes after deployment. Continual learning (CL) enables this adaptation without retraining from scratch, yet applying it to recurrent sEMG models is not straightforward. Standard recurrent training practices can destabilize learning, and which CL strategy holds up under realistic cross-user drift is unestablished. We evaluate three representative CL strategies under domain- and class-incremental drift on NinaPro DB1, and find experience replay to outperform elastic weight consolidation (EWC) and naive fine-tuning. It reaches 83.3% final accuracy for class-incremental drift while limiting forgetting to 1.8-5.8% accuracy drop, as against 16.2-28.5% for EWC and naive training. We further show these models must be trained on shuffled, independently-windowed segments. Streaming windows in temporal order collapses the network into predicting a single dominant gesture, while a shuffled, stateless regime trains stably. Finally, hand-crafted time-domain features raise per-task accuracy by up to about 40% over raw signals, but only replay preserves that gain under class-incremental drift.