A generative physical model approach for enhancing the stroke palette for robotic drummers

The main technical contribution of this work is a generative model for stroke generation in robotic drummers based on the physics of the interaction between human hand and a drumstick. A wide palette of strokes such as multiple bounce, double and triple strokes can be generated using the same model by varying one single parameter, the time-varying torque applied by the thumb on the stick. We have also developed a number of post-processing tools have been developed to further modify the strokes produced by the physical model in an effort to simulate and capture the acoustic variety and richness of natural human drumming. We present results from a pilot study in which 22 subjects participated in a listening test in which they listened to 10 pairs of audio files, each consisting of a human and a robotic drummer playing multiple bounce strokes. Inferential confidence interval approach was used to determine the maximum probable difference between the data obtained from the listening test and a simulated experiment in which the robotic drummer was perceptually indistinguishable from a human drummer. The results indicate that this generative physics model marks an important step toward achieving human-like musical expressivity in robotic drummers.