Abstract
Exploration is a critical for adaptation, such as when learning a new tennis stroke. Past work has shown greater exploration along task dimensions that have little bearing on task success (task-irrelevant) compared to dimensions that impact task success (task-relevant). We do not know the distinct roles of reward-based or error-based processes during exploration. Here we test the idea that reward feedback and error feedback differentially impact exploration. In Experiment 1, we predicted that reward feedback would lead to greater exploration in the task-irrelevant dimension. In Experiment 2, we predicted less exploration when participants received error feedback compared to reward feedback. To investigate, participants grasped the handle of a robotic manipulandum and reached to a target without vision of their hand. With reward feedback, participants heard a pleasant sound when their final hand position (endpoint) hit the target. With error feedback, participants saw a cursor representing their endpoint position. We calculated trial-by-trial endpoint autocorrelations to quantify exploration. In Experiment 1, participants received reward feedback when they reached to a short (task-relevant) or long (task-irrelevant) rectangular target. Participants displayed greater exploration when reaching to the task-irrelevant target (p = 0.005). In Experiment 2, participants reached to a long rectangular target and received either reward feedback, error feedback, or both forms of feedback. Compared to reward feedback, participants explored less with error feedback (p = 0.001) and both forms of feedback (p = 0.007). Collectively these preliminary results suggest that reward feedback is critical to exploration and is suppressed by error feedback.
