RoboPianist: Dexterous Piano Playing with Deep Reinforcement Learning

TL;DR

RoboPianist demonstrates how deep reinforcement learning can enable simulated robot hands to master complex piano playing, achieving high dexterity and coordination across 150 pieces, with open resources for future research.

Contribution

The paper introduces RoboPianist, a novel deep RL system for high-dimensional dexterous control in piano playing, along with an open environment, benchmarks, and evaluation tools.

Findings

Successfully learned to play 150 piano pieces with simulated hands

Provided open-source environment, datasets, and benchmarks for future research

Achieved high spatial and temporal precision in complex finger coordination

Abstract

Replicating human-like dexterity in robot hands represents one of the largest open problems in robotics. Reinforcement learning is a promising approach that has achieved impressive progress in the last few years; however, the class of problems it has typically addressed corresponds to a rather narrow definition of dexterity as compared to human capabilities. To address this gap, we investigate piano-playing, a skill that challenges even the human limits of dexterity, as a means to test high-dimensional control, and which requires high spatial and temporal precision, and complex finger coordination and planning. We introduce RoboPianist, a system that enables simulated anthropomorphic hands to learn an extensive repertoire of 150 piano pieces where traditional model-based optimization struggles. We additionally introduce an open-sourced environment, benchmark of tasks, interpretable evaluation metrics, and open challenges for future study. Our website featuring videos, code, and datasets is available at https://kzakka.com/robopianist/