Visual Feedback Is Necessary for Calibrating an Internal Model During Learning of a Novel Sensorimotor Task

Abstract

Optimal sensorimotor control is a complex task that is beautifully performed by the Central Nervous System (CNS). A great part of our control scheme is related to our capacity of learning novel skills. Theories of sensorimotor control have proposed the use of internal models by the CNS as a way to improve control when facing uncertainties, perturbations and to overcome slow feedback. In this paper, we investigated the dynamics of learning an internal model of a novel sensorimotor task during trials with and without visual feedback. In order to do so, 5 participants performed a sensorimotor task that translated force into positions of a cursor on the screen. We observed that performance gradually improved during training in both sensory modalities (correlation between learning curves was 0.89). However, the absence of visual feedback is detrimental to performance and somatosensory feedback alone was not sufficient to match performance when visual feedback is available. This suggests that error-signals given in visual coordinates are necessary to calibrate the internal model being learned.