Visuomotor memory is not bound to visual motion.

The motor system adapts its output in response to experienced errors to maintain effective movement in a dynamic environment. This learning is thought to utilize sensory prediction errors, the discrepancy between predicted and observed sensory feedback, to update internal models that map motor outputs to sensory states. However, it remains unclear sensory information is relevant (e.g., the extent to which sensory predictions depend on visual feedback features). We explored this topic by measuring the transfer of visuomotor adaptation across two contexts where input movements created visual motion in opposite directions by either: (i) translating a cursor across a static environment or (ii) causing the environment to move towards a static cursor (272 participants: 94 male, 175 female). We hypothesized that this difference in visual feedback should engage distinct internal models, resulting in poor transfer of learning between contexts. Instead, we found nearly complete transfer of learning across contexts, with evidence that the motor memory was bound to the planned displacement of the hand rather than visual features of the task space. Our results suggest that internal model adaptation is not tied to the exact nature of the sensory feedback that results from movement. Instead, adaptation relies on representations of planned movements, allowing a common internal model to be employed across different visual contexts. Human motor control requires constant calibration to remain effective in a dynamic environment. This adaptive process is thought to be driven by error-based learning in internal models that either predict the sensory consequences of a planned movement or output the required movement to realize a sensory goal. However, what sensory information is relevant is unclear. We probed whether internal model adaptation, in response to rotated visual feedback, transferred across two contexts where a common hand movement caused visual motion in opposite directions. We found near-complete transfer of learning across these two contexts, and that learning was tied to hand movements. These results indicate that internal models operate at a level abstracted from the exact nature of the visual feedback provided.