Isolating error-based and reward-based learning in a real-world task via gradual perturbations in embodied VR.

Error-based and reward-based mechanisms act together in real-world motor learning. Laboratory tasks separate them by manipulating feedback, but we previously showed that in a real-world task the correction of a large error is rewarding in itself, and engages reward-based learning even without a reward signal. Here, we asked whether they separate when errors are kept small. We used embodied virtual reality of pool billiards, in which the visual scene is aligned with a physical table, and rotated the cue ball's seen trajectory gradually while haptics and proprioception stayed unchanged. Thirty-two participants played two sessions, learning the same rotation with error-only feedback in one and reward-only feedback in the other. Their exploration after failed shots and their post-movement beta rebound over the motor cortex differed between sessions. Gradual perturbations can therefore isolate the two mechanisms in a real-world task, and open a way to target them in rehabilitation and skill training.