Automatic Determination of Bowing Movement for Violin-playing Robot

—Differences in Total Rewards Depending on the Musical Piece and Tempo—

Abstract

Playing the violin is one of the important methods for expressing emotion for robots. In this study, we focused on the violin, and aimed to develop a system that automatically determines the bow movement from musical scores. We adopted Q-learning and ε-greedy method with a value function approximation by neural network. We adopted a different formula for calculating the rewards from our previous studies. We selected three musical pieces and several BPM values. As a result of the simulation, for all the musical pieces, there were more positive rewards than negative ones.