Reproduce “UNITAS: A 3D-Native World Action Model for Embodied Manipulation” 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
Ruixiang Wang, Yongyi Su, Wenlve Zhou, Bo Yue, Hengyan Liu, Dekun Lu, Yuxin Tian, Yihan Fang and 9 more
UNITAS models manipulation in a calibrated metric 3D frame, representing observations, gripper motion, and scene-point responses within each prediction window. It encodes gripper trajectories as action flow and scene displacements as scene flow; a separate branch generates robot commands. On RoboTwin, Moving ADE is 0.427 cm for UNITAS versus 0.656 cm for PointWorld under matched motion conditions, measuring scene prediction with supplied motion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.