強化学習を用いたバイオリン演奏ロボットの動作決定

―学習条件と音圧の変更による演奏動作の相違―

抄録

We are building a system that can automatically determine the bow movement including the bow speed and the bowing direction from the musical score for the violin playing robot we built using the reinforcement learning. This report describes the effects of parameters including the number of hidden layers, the number of units in the layers, and the target sound pressure of each note in a musical score. We adopted Q-learning as the learning strategy and a neural network as the value function. We conducted simulation experiments and analyzed the effects of the parameters on the output bow movement. As a result, we found better parameter values. We also confirmed that the system output the bow movement using different sound pressure pattern.