Long-term motion generation for interactive humanoid robots using GAN with convolutional network

Abstract

In this report, we propose a framework for generating long-term human-like motion based on a deep generative model. Thanks to the network structure, the proposed method allows us generating seem-less long-term motions while the model is trained by 4 seconds long short motion samples. The computer graphics of generated motions seem to be reproduced scenes where a pair of persons talking to each other.