Apple published a new study called EMBridge demonstrating how to train AI recognizing hand gestures not included in its original dataset. The research, to be presented at ICLR 2026 Conference in April, shows how the AI model interprets muscle signals from wearable devices. This breakthrough could enable new interaction methods for future Apple products.

The researchers developed a cross-modal framework bridging EMG signals and hand pose data. Using datasets with 370 hours of sEMG across 193 users, the model achieved zero-shot gesture recognition with only 40% of training data. This approach allows the system to generalize to gestures it had never encountered before.

While the study doesn't mention specific products, the technology holds potential for wearable Human-Computer Interaction. This could include controlling devices like Apple Vision Pro through hand gestures detected by wearables. The research establishes Apple's advancement in EMG-based interaction, positioning them alongside competitors like Meta exploring similar technology.