A Learning Method for Stiffness Control of a Drum Robot for Rebounding Double Strokes

Abstract

In robot drumming, performing double stroke rolls is a key ability. Human drummers learn to play double strokes by just trying it several times. For performing it, a model needs to be learned to provide anticipatory commands during drumming. Joint stiffness plays a key role in rebounding double stroke task and should be considered in the model. We have introduced an interactive learning method for a drum robot to learn joint stiffness for rebounding double stroke task. The model is simulated for a 2-DoF robotic arm. The algorithm is simulated with 3 different drum kits to show the robustness of the learning approach. The simulation results also show significant compatibility with human performance results. In addition, the refined learning algorithm adjusts the stroke timing which is important for producing proper rhythms.