De Novo Sensorimotor Learning Through Reuse of Movement Components

Abstract

AbstractFrom tying one’s shoelaces to driving a car, complex skills involving the coordination of multiple muscles are common in everyday life; yet relatively little is known about how these skills are learned. Recent studies have shown that new sensorimotor skills involving re-mapping familiar body movements to unfamiliar outputs cannot be learned by adjusting pre-existing controllers, and that new task-specific controllers must instead be learned “de novo”. To date, however, few studies have investigated de novo learning in scenarios requiring continuous and coordinated control of relatively unpractised body movements. In this study, we used a myoelectric interface to investigate how a novel controller is learned when the task involves an unpractised combination of relatively untrained continuous muscle contractions. Over five sessions on five consecutive days, participants learned to trace a series of trajectories using a computer cursor controlled by the activation of two muscles. The timing of the generated cursor trajectory and its shape relative to the target improved for conditions trained with post-trial visual feedback. Improvements in timing transferred to all untrained conditions, but improvements in shape transferred only to untrained conditions requiring the trained order of muscle activation. All muscle outputs in the final session could already be generated during the first session, suggesting that participants learned the new task by improving the selection of existing motor commands. These results show that generating novel motor behaviours need not involve learning to generate new motor commands.Significance StatementReal-world skills often involve continuous coordination of multiple muscles. Such skills cannot be learned by adjusting an existing control policy, instead requiring a new controller to be learned “de novo”. It remains unclear how new controllers are learned for tasks involving unfamiliar combinations of body movements. In this study, we used a novel human-computer interface task to test this. Over five sessions, participants learned to trace a series of cursor trajectories by coordinating the activation of two muscles. We found that participants tended to reuse the same muscle contractions for trained and untrained variants of the task, and that performance improvements were attributable to improvements in the choice of muscle contractions from a pre-existing repertoire. Our results demonstrate that learning of new complex movements does not necessarily require learning to generate new patterns of muscle activity.