Reproduce “PANDORA: Diffusion Policy Learning for Dexterous Robotic Piano Playing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yanjia Huang, Renjie Li, Zhengzhong Tu
Researchers from Texas A&M University introduce PANDORA, a diffusion-based framework for robotic piano playing that combines conditional U-Net architecture with LLM-driven semantic rewards, achieving state-of-the-art performance in both technical precision and musical expressiveness while reducing training time compared to existing approaches.
Key insights
- LLM-driven semantic rewards can effectively guide learning towards more expressive and human-like performance
- Combining diffusion models with residual refinement enables both smooth trajectories and precise execution
- Web-scraped professional piano performances provide rich training data when processed with MediaPipe and DepthAnything
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.