Reproduce “Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging” 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

Yuanchu Liang, Yue Yang, An-Chi He, Hanna Kurniawati

LARBRID fine-tunes a pretrained diffusion policy by optimizing its initial-noise distribution while keeping the generator fixed. It learns a shared representation that aligns value estimates for each noise input and its decoded action, addressing noise aliasing. In simulated Aloha cube transfer, it achieves twice DSRL’s success rate at 50k–80k training steps; this comparison is limited to early training.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.