Human motion imitation for humanoid by Recurrent Neural Network

Abstract

In this paper, it is investigated how a humanoid can be controlled from a human motion in a tele-operation system. The proposed method is based on Recurrent Neural Networks (RNN) to extract features in nonlinear sequential data. Therefore, our proposed RNN model can extract the features of the relation between a human and a robot, and generate motion of a robot using these features. Also, the generated robot motion is modified to maintain balance with stable contact on the ground. The effectiveness of the proposed new mapping algorithm is verified through simulations.