Perceptual sensitivity supports online control, while perceptual errors drive motor memory formation during locomotor adaptation

To maintain stable locomotion, the nervous system must continually adapt, whether reacting to an unexpected perturbation, such as a trip on uneven terrain, or anticipating external demands, such as walking on snow, by forming and updating motor memories. Error-based learning is the dominant computational account of such adaptation, in which motor commands are updated to reduce prediction errors, that is, the mismatch between predicted and actual limb state. Yet this framework was largely defined in reduced, single-effector paradigms, where the sensory consequences of movement are isolated and the prediction error is directly observable. Whether the same principle governs whole-body, multi-segmental, multi-sensory behaviors such as walking has remained untested, owing in part to the challenge of identifying a behavioral proxy for prediction errors in this complex, dynamic setting. Here, we combined a split-belt treadmill paradigm with a novel method for quantifying perception of leg motion to address this open question. We found two perceptual contributions to locomotor adaptation: perceptual sensitivity predicted initial motor corrections during early adaptation, whereas perceptual errors (i.e., the mismatch between perceived and observed leg speed) predicted the magnitude of motor aftereffects during post-adaptation. Together, these findings demonstrate that locomotor adaptation is fundamentally constrained by perception of limb motion. Our results identify perceived limb motion as a behavioral proxy for prediction errors, establishing error-based learning, long characterized in reduced tasks, as a core computational principle underlying human locomotion.