Intrinsic and extrinsic noise in motor skill acquisition

Abstract

Research on human motor skill acquisition has typically examined mean and variance of performance metrics to evaluate learning processes. Skill has been equated with low variability to be reduced with practice. Continuing recent efforts, this thesis focused on the structure of movement variability, both in its distribution and temporal evolution, as a window to understand motor learning and control. While random components are assumed to arise from intrinsic and extrinsic noise, some of the overt behavioral variability is also structured and purposefully generated by the central nervous system. In this thesis motor skill acquisition is viewed as the strategic search for task solutions where intrinsic and extrinsic noise minimally impact task performance. Eight studies examined the role of noise in motor learning and control and used a target-oriented throwing task as testbed. Importantly, the task has redundancy with a solution manifold defining the set of solutions that achieve task success. The throwing task was experimentally controlled, modeled and rendered in a virtual environment to allow mathematically-based manipulations of the solution space to address detailed questions on the role of noise in motor control. Chapter 2 identified human strategies in task variations with different geometries of the result space showing that subjects' throwing movements exploited the task's redundancy to reduce the sensitivity of the ball release time to intrinsic noise. Chapter 3 revealed that when subject performed a sequence of discrete throwing movements, they merged the discrete actions into a rhythmic sequence that displayed stability, i.e. robustness with respect to noise, and achieved better task success. Based on these empirical and theoretical findings, two subsequent studies (Chapter 4-5) created interventions in the virtual environment that assisted subjects to suppress their noise and achieve better vi performance. System identification with stochastic state space models revealed that visual amplification of performance error led subjects to suppress their noise with extended practice. Adding noise to the time of ball release exerted "pressure" that lowered the timing errors and thereby improved task performance. Importantly, this benefit persisted even after removal of the manipulation. Chapter 6 explored how the central nervous system coordinates the variability in the multi-dimensional space of muscle activations. Using dimensionality reduction methods, the data revealed four muscle synergies that were invariant across subjects and sequentially supported the multi-joint throwing action. Chapter 7 and 8 tested the robustness of these finding in the virtual environment in a real implementation and in a "field study". The comparison between virtual and real performance showed that, even though mean performance in the real setup was better, the variability was modulated across practice in the same manner in both environments. Deploying a portable setup at the Museum of Science in Boston and testing a large coho...