Instrumental-Motor Transfer: The Relative Value of Competing Movement Goals Modulates Implicit Motor Learning

Abstract

Motor learning depends on the integration of several, often competing signals, including reward and punishment (which guide learning the value of actions via reinforcement learning) and sensorimotor errors (which refine the execution of actions via motor adaptation). If and how these signals interact remains controversial. Here we show that the learned value of actions can fundamentally alter how the motor adaptation system learns from errors. Human participants learned the reward values of different movement targets and subsequently underwent implicit motor learning at those targets. Across several experiments, motor learning was robustly attenuated when it brought the limb toward previously punished movement directions. Moreover, learning was globally suppressed when punished targets were themselves the movement goal. Learning was not reciprocally enhanced when it pushed movements toward previously rewarded directions, revealing a valence asymmetry. Using computational modeling, we present a novel mechanism to explain our effects: the motor adaptation system tunes its error sensitivity according to a movement-specific value gradient. Our findings point to a close interaction between reinforcement learning and motor memory -- a kind of "instrumental-motor transfer" -- where learned action values modulate error-based motor adaptation.