| Throughout life, we learn and perform movements in the presence of our intrinsic movement variability. With absolutely zero variability, the motor system would not be flexible enough to explore the world around us, and yet if the variability is too high, the system could become unstable. We are constantly dealing with this balance between movement variability and task performance. However, motor control and learning theories seldom address this relationship. This interaction between movement variability and motor learning should not be ignored in motor learning analysis, especially in populations with high movement variability, like children with secondary dystonia. Dystonic children have increased variability in the arm for both reaching movements and single joint isometric force production. Through this thesis, we hope to better understand the relationship between variability and motor learning, especially in children with dystonia. In this work, we used experiments that mapped arm kinematics to two computerized ball games to study the how dystonic subjects respond to perceived variability, and use variability in learning.;We investigated the effects of increased and decreased variability on motor learning. We performed simulations of systems that respond to variability and we studied how normal and dystonic subjects responded to increased and reduced perceived variability by manipulating the feedback display. Results showed that both normal and dystonic subjects chose strategies that reflect the changes in perceived variability. Then, we demonstrated a framework to separate the cost component related to the variability from the other components in motor tasks. This decomposition of performance allows us to examine two aspects of improvement: (1) exploration to find the correct target, and (2) reduction in variability to reduce cost near that target. The results suggest that reduction in variability is more important for subjects with high baseline variability.;Furthermore, we investigated the learning algorithms used by the subjects. We propose that noise can potentially be a contributor to learning, such as a cost dependent noise algorithm (CDN). The CDN algorithm does not require an explicit knowledge of the cost gradient, which is not always readily available. The experimental results showed that subjects likely used a CDN like algorithm during learning, in combination with other deterministic algorithms such as gradient descent. Careful examination of the CDN algorithm revealed that though this algorithm could be used with limited restrictions, the algorithm might not converge when the noise was too high. Therefore, there is a delicate balance between noise that helps learning and noise that hinder convergence. Since noise (CDN in particular) was potentially useful in learning in certain circumstances, we hypothesized that adding exogenous CDN could improve learning. We tested the hypothesis by performing an experiment where training in an environment with exogenous CDN and tested the post training performance and retention on the next day.;In summary, our results showed that subjects, both normal and dystonic, optimized their motor strategies based on perceived motor variability. The dystonic patient's abnormal movements may be due to the increased noise causing them to choose solutions that would not otherwise have been optimal. Also, we showed that subjects may use noise and variability as a learning algorithm, such as in CDN. Furthermore, externally adding CDN could potentially be useful as a training paradigm for patients with dystonia and other movement disorders, warranting further studies. Based on our findings, we conclude that variability plays a large role in motor learning, as it affects planning and learning. |
