Reproduce “Environment Edits as a Corrective Signal for Reward Learning” 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
Isaac Ngui, Courtney McBeth, Nancy M. Amato
FERRET adds scene edits to robot corrections: people move objects while robot motion stays fixed. Its models represent object relations and attributes. In an ambiguous laptop-and-cup test, robot corrections failed to identify the laptop, but environment traces let the relational model learn a feature similar to ground truth. The evidence is from three author-generated scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.