筋骨格モデルを用いた運動学習におけるモジュラリティの役割検証

抄録

Clarification of how our brain and central nervous system (CNS) control the human body is important for further development of engineering fields and medical fields. Although it remains an open problem due to its complexity, there is a hypothesis that our CNS controls modules which handles tasks or dynamics by activating several muscles synergically rather than individual muscles. This hypothesis is called “Muscle Synergy Hypothesis” and currently discussed through abundant physiological experiments. However, there are not enough studies those focus on what is the purpose or advantage of the existence of such a modular controlling nervous structure up to now. Here, we aim to investigate the role of modularity of nervous control system regarding motor learning. For the purpose, we constructed two kind of motor learning models those consist of musculoskeletal model and neural network: the one has modular structure and the other does not. After that we trained these models and compared learning performance. Besides, we also tested reusability of modules among tasks. The results showed that the model with modules performed faster and more accurate motor learning and stronger generalization than the model without modules. Moreover, we found that modules which is obtained under certain muscle dynamics can be reused for another muscle dynamics.