Reproduce “Cleanly-DINO: Efficiently Learning Visual Task Spaces for Robotics” 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

Adrian Röfer, Melika Shekarriz, José Arce y de la Borbolla, Iman Nematollahi, Abhinav Valada

Cleanly-DINO learns patch-level visual progress for robot cleaning, locating regions that remain dirty rather than assigning progress to a whole image. It uses masked video patches, frozen DINOv2 features, and a learned projection with a filter for task-aligned changes. On held-out demonstrations, it reaches 0.763 ± 0.07 AUROC versus 0.721 ± 0.08 for raw DINOv2, within tested cleaning settings.

Summary by alphaXiv.

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