Al-Fawakhiri, N.; Chib, V. S.; McDougle, S.

Multiple learning signals can shape motor output, including reward and punishment (via value-based reinforcement learning) and sensorimotor error (via motor adaptation). However, it is unclear if action values, learned via reinforcement learning, interact with error-based motor learning. Here, we asked if the learned value of competing motor goals influences how the motor system learns from sensorimotor errors linked to those goals. We designed a paradigm that required participants (N = 85) to learn about the value associated with different movement targets prior to undergoing implicit visuomotor adaptation at or nearby those same targets. We observed two classes of related effects: repulsion and suppression. When adaptation brought the limb toward targets associated with low-value, learning was attenuated compared to when adaptation brought the limb away from targets associated with low-value, indicating a "repelling" effect of low-value actions. Moreover, adaptation was suppressed in all directions when low-value targets were themselves the goal of movement. These findings were asymmetric with respect to valence: we did not observe comparable attraction effects when adaptation brought the limb toward targets associated with high-value, nor did we see overall enhancement of adaptation when targets associated with high-value acted as movement goals. Additional analyses and experiments demonstrated that these effects did not reflect generic biases and were driven by relative rather than absolute goal values. Repulsion and suppression effects were tied to movement directions rather than the targets themselves. These results point to a novel interaction between reinforcement learning and motor memory - a kind of "instrumental-motor transfer."