Abstract
Humans can adapt to large and sudden perturbations of sensory feedback. What multisensory and motor-execution cues are used to determine confidence in action success, and do the dynamics of confidence parallel those of ongoing sensorimotor adaptation? Participants made a slicing reach through a visual target with an unseen hand, followed by a continuous judgment of confidence in reach success. For the confidence judgment, participants adjusted the size of an arc centered on the target. Larger arcs reflected lower confidence. Points were awarded if the subsequent visual feedback was within the arc, and fewer points returned for larger arcs. This incentivized attentive reporting of confidence and minimizing feedback-target distance to maximize the score. After the confidence response, visual feedback of hand position was shown at the same distance along the reach as the target. A 20 deg rotation was applied to the feedback during the central 50 trials of a block (alternating between clockwise and counterclockwise across blocks). We used least-squares cross validation to compare four Bayesian-inference models of sensorimotor confidence using prospective cues (knowledge of motor noise and visual feedback from past performance), retrospective cues (proprioceptive measurements), or both sources of information integrated to maximize expected gain (an ideal observer) with additional parameters for learning and bias. All but one of the participants used proprioception to calculate sensorimotor confidence during motor adaptation in addition to prior information. Confidence recovered exponentially to pre-adaptation levels after the perturbation ended, but at a slower rate than motor learning.