emg2tendon

Mapping surface EMG to tendon control forces for dexterous, musculoskeletal robotic hands (RSS 2025).

From Muscle Signals to Tendon Control

Tendon-driven robotic hands are far more dexterous than joint-actuated designs, but commanding them is hard: motion-capture data doesn’t map cleanly onto tendon controls, and vision-based tracking fails under occlusion. Surface electromyography (sEMG) — read from a wrist-worn band — is a robust, inexpensive alternative, but mapping raw sEMG to tendon control forces is an open problem.

emg2tendon (independent research, RSS 2025) attacks exactly that mapping. sEMG recordings are paired with tendon-control targets derived from the MyoSuite MyoHand musculoskeletal model, and regression models — including a novel diffusion-based approach — predict tendon forces from muscle activity.

Contributions

  • EMG-to-Tendon Control dataset extending emg2pose with tendon annotations: 193 subjects, 370 hours, 29 gesture categories.
  • Three baseline regression models plus a diffusion-based predictor.
  • A C++ QForce implementation with a ~20× speedup, processing 1,700 hours of mocap in 24 hours.

Project page: emg2tendon.github.io · Read the blog post. </content>