The brain uses invariant dynamics to generalize outputs across movements

Abstract

The nervous system uses a repertoire of outputs to produce diverse movements. Thus, the brain must solve how to issue and transition the same outputs in different movements. A recent proposal states that network connectivity constrains the transitions of neural activity to follow invariant rules across different movements, which we term ‘invariant dynamics’. However, it is unknown whether invariant dynamics are actually used to drive and generalize outputs across movements, and what advantage they provide for controlling movement. Using a brain-machine interface that transformed motor cortex activity into outputs for a neuroprosthetic cursor, we discovered that the same output is issued by different activity patterns in different movements. These distinct patterns then transition according to a model of invariant dynamics, leading to patterns that drive distinct future outputs. Optimal control theory revealed this use of invariant dynamics reduces the feedback input needed to control movement. Our results demonstrate that the brain uses invariant dynamics to generalize outputs across movements.