Mapping Grasping Motion of Hand between Master and Slave in a Low-Dimensional Latent Space

In recent years, the demand for master-slave robots capable of performing various tasks has been increasing. However, mapping the motion of a master human hand to a slave robot hand is difficult for two reasons. The first reason is the different kinematic structures of the human and robot hands, which leads to a different dimensionality representation of motions of the hands. The second reason is the difficulty of modeling the hand's motion; since the natural motion of the human hand is complex, an accurate representation of it has high dimensionality. In recent studies, it has been shown that some specific commonly used motions, such as handling and grasping of objects, can be represented in a lower-dimensional non-linear manifold in hand posture space. In this study, considering the problem mentioned above, we propose a method for modeling hand motion in a low-dimensional latent space by using Gaussian Process Latent Variable Models (GPLVMs), and learning the relationship between the grasping motion of master hand and the slave hand by using Gaussian Mixture Regression (GMR).