Experiments on Imitation Learning of Humanoid Robot with Deep Reinforcement Learning and a Consideration on Symbol Grounding Problem

Abstract

Imitation of behavior is one of the important elements of the development of intelligence, as can be seen from the fact that human babies grow up while imitating the behavior of the surrounding human beings. Such imitation learning of behavior was not easy because robots were expensive and algorithms were complicated. However, recently humanoid robots have become inexpensive, and the progress of deep learning has made algorithms development easier. Therefore, experiments were applied to imitation learning of behavior by humanoid robots, and the results are reported. The relationship between this method and the symbol grounding problem also be described.