Representation Learning / Motor Control
Perturbation Recovery CVAE
Learning whether early neuromuscular signals encode transferable control policies across humans.
Details
Built a CVAE using early EMG signals from 0-80ms post-perturbation on the Lorenz treadmill perturbation dataset. The latent representation predicts cross-subject muscle synergy activations while direct mechanics prediction fails.
Why it matters
Suggests control structure can transfer across humans while mechanical execution remains body-specific.