Recently, robots are expected to work in real environment to support our daily lives. However, there are a lot of disturbances in such a environment. A robot with many degrees of freedom might be beneficial in coping with various disturbances in a real environment, because it can generate many kinds of motion. However, because of its complicated structure, it is difficult to estimate its dynamics which is necessary to design the control rule. In this paper, we discuss about 3 data-driven methods (k-nearest neighbor, support vector regression and Gaussian process regression) for estimation of a human-like upper body musuculoskeletal robot driven by air actuators.
Kenji Urai (浦井 健次) >