Reproduce “GR-VIP: Geometric representation conditioning for precise visuomotor policy” 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

Chanhwi Lee, Jiwon Seong, Eunje Choi, Hohyung Lee, Jongsoo Lee, Junyeong Bae, Inveom Kwak, Woosang Shin and 1 more

GR-VIP learns a robot’s docking corrections by comparing recent camera views with a goal image using dense cross-view features from a frozen Reloc3r encoder, then conditioning a diffusion policy on those features and wheel history. In an unseen room with a different station and background, it docked in 15/16 trials without additional training, a result limited to the tested setting.

Summary by alphaXiv.

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