Self-generated vestibular prosthetic input updates forward internal model of self-motion

The brain must differentiate between externally-generated and self-generated sensory inputs to build stable perception and generate appropriate behavior. Despite recent advances in neuroprostheses, little is known about how the brain interprets prosthetic sensory input, especially when self-generated. Here, we asked: How does the brain learn to distinguish self-generated prosthetic stimulation? To do this, we leveraged the well-understood neural pathways and behavior readouts of the vestibular system. A monkey was implanted with a vestibular prosthesis that restores sensory information via nerve stimulation and trained to make coordinated eye-head gaze-shifts between horizontal targets. Each session comprised a three-block learning paradigm: baseline gaze-shifts, gaze-shifts with prosthetic stimulation, and washout without stimulation. We hypothesized that 1) prosthetic stimulation would first engage vestibular reflex pathways, resulting in truncated gaze-shifts but also that 2) the brain would then update its internal model to account for this new sensory input. Consistent with our predictions, gaze position error initially increased after stimulation onset. Furthermore, this error then exponentially decayed within ~60 trials, with early washout trials showing oppositely-directed gaze position error, indicating a central adaptation rather than simply reflex suppression. We next asked whether the observed adaptation was due to updating a forward internal model (i.e., the brain’s prediction of self-generated prosthetic sensory input) or instead was simply the result of updating the gaze controller’s inverse model (i.e., the required motor command). To do this, we tested learning in a gaze adaptation paradigm where the target was jumped forward 5. This learning, which requires mainly the updating of inverse model, demonstrated substantially slower adaptation (~300 trials). Together, our results show that the brain can quickly learn to use self-generated prosthetic input by updating its forward internal model of self-motion. Importantly, these findings provide new insights on how the brain interprets prosthetic sensory inputs to generate accurate behavior.