HandelBot: Real-World Piano Playing via Fast Adaptation of Dexterous Robot Policies

Official implementation of HandelBot.

[Paper] [Website]

<img src="images/fig1.png" alt="HandelBot Method" title="HandelBot Method">

Installation

You can create the conda environment directly from the env.yml file:

conda env create -f env.yml
conda activate handel

Training in Simulation

We train with a custom ManiSkill piano environment. To train a policy:

python -m rl.piano_ppo_fast --horizon HORIZON --note_trajectory music_data/SONG_NAME.pkl --exp_name YOUR_EXP_NAME

To reproduce our results, we used the following horizon lengths:

• Twinkle Twinkle: 160
• Ode to Joy: 330
• Hot Cross Buns: 160
• Prelude in C: 330
• Fur Elise: 320

Real-World Training

Refer to real/src/dg5f_driver/README.md for instructions on running on the real robot.

Citation

@misc{xie2026handelbotrealworldpianoplaying,
      title={HandelBot: Real-World Piano Playing via Fast Adaptation of Dexterous Robot Policies}, 
      author={Amber Xie and Haozhi Qi and Dorsa Sadigh},
      year={2026},
      eprint={2603.12243},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2603.12243}, 
}