DEVELOPING IMITATION LEARNING TO PRODUCE RHYTHMIC MOTION OF DRUMMING IN NAO HUMANOID ROBOT USING SUPERELLIPSES

This paper discussed the development of imitation learning in Nao humanoid robot inspired by central pattern generator. Purpose of imitation learning is to enable people to teach new skills to robots via showing examples to them. In sum, the issue studied in this paper was imitation learning for rhythmic actions. Basics of robot learning in this paper included appropriate demonstration techniques, understanding teachers' demonstrations, and an introduction to reinforcement learning. The proposed method was formed based on the evolution of past techniques and superellipses were used to extend the application of imitation learning in drumming. Superellipses, as a wide range of closed geometric curves, are very important for engineers, because they have helpful and precise features in terms of design, programming, and control. The advantage of the proposed method was in the ability to encode learned periodic signals as a limit cycle of superellipses, which influenced louder or lower drumming. In fact, using these curves, an approximate solution was found for mapping line graphs to the desirable ones. Simulation results of the considered method were shown in Webots software and its implementation was demonstrated on a real Nao.