RoboPianist: Dexterous Piano Playing with Deep Reinforcement Learning

Pretty mid paper, mostly engineering.

“We observe that training on an increasing amount of songs is significantly harder than training specialist policies on individual songs”.

Citation

(Zakka et al., 2023)

Zakka, K., Wu, P., Smith, L., Gileadi, N., Howell, T., Peng, X. B., Singh, S., Tassa, Y., Florence, P., Zeng, A., & Abbeel, P. (2023). RoboPianist: Dexterous Piano Playing with Deep Reinforcement Learning. https://arxiv.org/abs/2304.04150