Reward prediction errors shape sensory-error-driven single-trial motor learning

Abstract

Short-term motor adaptation is a form of motor learning that optimizes how sensory information about a target’s location is translated into a target-directed movement, using sensory feedback on performance errors. When actual and predicted sensory feedback differ, sensory prediction errors (SPEs) are generated, enabling the cerebellum to adjust this mapping. In contrast, reinforcement learning uses reward prediction error (RPEs) based on fulfilled or unfulfilled reward expectations to shape future actions. As RPE-related information also reaches the cerebellum, we wondered whether it interacts with SPEs to shape short-term motor adaptation. We addressed this in monkeys by inducing visual errors via inward and outward intra-saccadic target steps, generating corresponding SPEs and quantifying adaptation as changes in saccade amplitude across same-direction trials before and after SPEs. In two separate studies, we manipulated information about trial outcomes (reward versus no reward) to examine its interaction with SPEs in saccadic adaptation. In both, we found that RPEs modulate single-trial saccadic adaptation. Sensory errors determined the direction of adaptation, whereas reward-related signals scaled its magnitude, revealing an interaction between SPEs and RPEs in trial-by-trial motor learning and underscoring the behavioural relevance of their co-representation in shared cerebellar afferents.