Effect of exploratory perturbation on the formation of kinematic synergies in Goal Babbling

Motor synergies explain how humans deal with control of their highly-redundant motor apparatus. In this study, we analyze the formation of motor synergies in a computational model for online motor learning, which is named Goal Babbling. The recently proposed Goal Babbling scheme addresses the learning of inverse models for motor control by goal-directed exploration and solves the redundancy problem by preferring efficient motions in the learning process. We show experimental results that verify the formation of kinematic motor synergies in the beginning of Goal Babbling. We further extract characteristics of the formed synergies under various conditions of exploratory perturbations. It turns out that the learned synergies strongly depend on the exploratory perturbation in action space.