Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 02, 2018 - June 05, 2018
In recent years, the demand for robots capable of performing various works in dangerous environments is increasing. In such robots, master-slave control that allows an operator to make a situation determination is suitable. However for some tasks, only master-slave control is not enough, appropriate autonomous assist control is needed for this situation. The propose of this study is to develop a control algorithm to assist task operation on a master slave system. In this paper, we propose a method for predict the hand motion of operator during master-slave control. We first clustering the operator’s hand motion into 4 most used pre-shaped grasp pattern, each pattern is performed 10 times to obtain the grasp data, analysis with nonlinear dimensionality reduction techniques. Then EM algorithm is performed to fit a Gaussian mixed model on the latent space extracted from grasp data. As results, the fitted GMM model is used to predict the hand motion of operator in real-time, the proposed method is verified by grasp experiment.
