Human Action Imitation System Based on Nao Robot

Abstract

A human action imitation system based on Nao robot is proposed in this paper. A new strategy, combining Momentum with Stochastic gradient descent(SGD), is designed to accelerate the training process of Convolutional Neural Networks(CNN). The CNN is used to recognize the 2D coordinate of joints of a human body. This data is translate to Nao Robot and make it to imitate human action. The experiments show that the training rate is faster than traditional training method. The Robot can imitate the human action accurately.