AI model decodes whole-body monkey movements from brain signals

A team led by Mu-ming Poo at the Chinese Academy of Sciences has published a proof-of-concept study demonstrating whole-body motion decoding from intracranial neural signals in freely moving monkeys. The work, posted on arXiv (arXiv:2605.29355), overcomes previous limitations of constrained tasks by recording from distributed sensory- and motor-related cortical areas via epidural electrodes while synchronizing multi-view video motion capture. Using an autoregressive encoder-decoder model, the researchers learned a compact behavioral prior and conditioned it on neural activity to generate accurate, realistic three-dimensional body kinematics without imposing explicit physical constraints.

The key innovation lies in scaling neural-behavioral modeling to naturalistic, unconstrained behaviors. While prior motor decoding focused on constrained limb movements, this method reconstructs the full body—including torso, limbs, and head—in complex, spontaneous actions. The autoregressive model captures temporal dependencies in both movement and neural signals, enabling smooth and plausible predictions. The authors suggest this approach could eventually inform brain-machine interfaces for more natural prosthetic control or rehabilitation, though work remains to translate the findings to humans. The paper is currently under review for publication.