Feedback-Feedforward Dynamics Shape De Novo Motor Learning

Abstract Skill learning relies on both corrective feedback and predictive feedforward control, yet how these processes evolve and interact during the acquisition of a new motor skill remains unclear. We addressed this question by examining how humans learned a continuous visuomotor mirror reversal tracking task over five days. To dissociate feedback- and feedforward-related contributions, we combined frequency-based analyses with responses to brief cursor perturbations and measurements of the initial direction in point-to-point reaches. Learning exhibited both temporal and frequency-dependent structure. Early improvements were dominated by rapid online feedback corrections expressed primarily at low movement frequencies. In contrast, predictive feedforward control emerged more gradually and became increasingly important at higher frequencies, where delayed feedback is insufficient for accurate control. A recurrent neural network model, in which an adaptive cerebellar-like feedback controller provides a teaching signal to a recurrent cortical predictive controller, reproduced these key behavioral signatures and suggested a mechanistic account in which feedback-driven error signals shape the gradual emergence of predictive control. Together, these findings support a framework in which de novo motor learning arises through distinct but interacting feedback and feedforward processes that co-evolve across practice under the constraints imposed by sensorimotor delays. Competing Interest Statement The authors have declared no competing interest. Footnotes New analysis and results, New computational model, Supplemental files updated.