Optimal gain schedules for visuomotor skill training using error-augmented feedback

Abstract

Motor Learning is heavily governed by sensory feedback, and artificially enhancing feedback influences learning as seen in our previous works. This study provides a model-based approach in determining the optimal gain schedules for augmented feedback on a visuomotor learning task. Using Gaussian process regression, we modeled the phenomenological process of learning to operate a robot with visual rotation. We then used Pontryagin's minimum principle to achieve the optimal feedback gain schedules that yield the fastest learning, the highest post-training performance, and both at the same time. Our results reveal that the instantaneous error feedback should be doubled (×1.92) throughout the training if the fastest learning is desired. However if the highest post-training performance is desired along with the fastest learning, the feedback gain should be gradually varied from 1.92 to 1. This study explores a novel approach to optimize specific aspects of training for areas such as robotic-neuro-rehabilitation, teleoperation, sports coaching, and human-machine interactions.