Learning through Imitation and Reinforcement Learning: Toward the Acquisition of Painting Motions

Abstract

Learning is essential for an autonomous agent to adapt to an environment. One method of learning is through trial and error, however, this method is impractical in a complex environment because of the long learning time required by the agent. Therefore, guidelines are necessary in order to expedite the learning process in such environments, and imitation is one such guideline. Sakato, Ozeki, and Oka (2012-2013) recently proposed a computational model of imitation and autonomous behavior by which an agent can reduce its learning time through imitation. They evaluate the model in discrete and continuous spaces, and apply the model to a real robot in order to acquire painting skills. Their experimental results indicate that the model adapted to the experimental environment by imitation. In this paper, we introduce the model and discuss what are needed to improve the model.