Reproduce “Distilling Instructions into Weights: Closed-Form Edits Fix a Robot Policy's Misreading” 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
Ali-Adeeb Abbas, Esmaeil Seraj
The authors use a known-good rewording to repair a zero-shot robot policy without training or changing the instruction. They record the rewording’s internal effects and copy them into the policy’s weights with a closed-form update. On held-out simulator starts, success rose from 27/128 to 106/128. The result applies to one checkpoint and task, and requires a known-good wording.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.