The Consistency of Prior Movements Shapes Locomotor Use-Dependent Learning

Repetition is an indispensable component of motor skill acquisition. However, it is unknown how consistent repeated movement patterns must be to engage an implicit "use-dependent" learning mechanism. In this Registered Report, we tackled this question through a combination of computational modeling, simulations, and behavioral experiments involving visually-guided treadmill walking. Our hypotheses were formalized by two distinct computational models: in the two-process Strategy plus Use-Dependent model, use-dependent learning is viewed as a slowly updating and slowly decaying bias in the direction of repeated movements. The Adaptive Bayesian model frames use-dependent learning as an emergent property of quickly adapting prior probabilities of target step lengths. Critically, the Adaptive Bayesian model is much more sensitive to variable practice than the Strategy plus Use-Dependent model. To test these hypotheses, human participants (N = 18, 10 females) learned a novel asymmetric stepping pattern under three conditions with differing amounts of practice consistency during a learning block. We probed use-dependent movement biases immediately postlearning by asking participants to "walk normally" during a washout block with no visual feedback (VF). We found that the total magnitude of use-dependent learning depended on practice consistency during learning, consistent with the Adaptive Bayesian model. However, this dependence faded quickly as biases became similar in magnitude over subsequent strides across all conditions, an observation more consistent with the Strategy plus Use-Dependent model. Simple post hoc adjustments to the Strategy plus Use-Dependent model made clear that these seemingly opposing effects of practice consistency can result from a unitary use-dependent learning process shaped by recent movement history.