Variability in motor learning: relocating, channeling and reducing noise.

Variability in motor performance decreases with practice but is never entirely eliminated, due in part to inherent motor noise. The present study develops a method that quantifies how performers can shape their performance to minimize the effects of motor noise on the result of the movement. Adopting a statistical approach on sets of data, the method quantifies three components of variability (tolerance, noise, and covariation) as costs with respect to optimal performance. T-Cost quantifies how much the result could be improved if the location of the data were optimal, N-Cost compares actual results to results with optimal dispersion at the same location, and C-Cost represents how much improvement stands to be gained if the data covaried optimally. The TNC-Cost analysis is applied to examine the learning of a throwing task that participants practiced for 6 or 15 days. Using a virtual set-up, 15 participants threw a pendular projectile in a simulated concentric force field to hit a target. Two variables, angle and velocity at release, fully determined the projectile's trajectory and thereby the accuracy of the throw. The task is redundant and the successful solutions define a nonlinear manifold. Analysis of experimental results indicated that all three components were present and that all three decreased across practice. Changes in T-Cost were considerable at the beginning of practice; C-Cost and N-Cost diminished more slowly, with N-Cost remaining the highest. These results showed that performance variability can be reduced by three routes: by tuning tolerance, covariation and noise in execution. We speculate that by exploiting T-Cost and C-Cost, participants minimize the effects of inevitable intrinsic noise.