PianoMime: Learning a Generalist, Dexterous Piano Player from Internet Demonstrations
This addresses the challenge of learning dexterous robotic skills from large-scale, noisy internet data, though it is incremental in leveraging existing methods for policy distillation.
The authors tackled the problem of training a generalist piano-playing agent from internet demonstrations, achieving up to 56% F1 score on unseen songs.
In this work, we introduce PianoMime, a framework for training a piano-playing agent using internet demonstrations. The internet is a promising source of large-scale demonstrations for training our robot agents. In particular, for the case of piano-playing, Youtube is full of videos of professional pianists playing a wide myriad of songs. In our work, we leverage these demonstrations to learn a generalist piano-playing agent capable of playing any arbitrary song. Our framework is divided into three parts: a data preparation phase to extract the informative features from the Youtube videos, a policy learning phase to train song-specific expert policies from the demonstrations and a policy distillation phase to distil the policies into a single generalist agent. We explore different policy designs to represent the agent and evaluate the influence of the amount of training data on the generalization capability of the agent to novel songs not available in the dataset. We show that we are able to learn a policy with up to 56\% F1 score on unseen songs.