Reproduce “See like a Robot: Robot-Centric Pointmaps for VLA Models” 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
Byungkun Lee, Dongyoon Hwang, Dongjin Kim, Hojoon Lee, Hyunseung Kim, Jaegul Choo, Minho Park
SeeR-VLA represents depth as an image-aligned pointmap whose 3D coordinates use the end effector as origin and robot-base-aligned axes, then fuses these features with RGB features in a pretrained VLA. On four real-world tasks, it averages 70.0% success, versus 46.3% for PointVLA and 37.5% for RGB-only π0.5, under a tested training-viewpoint protocol, not held-out viewpoint generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.