Reproduce “UniAction: Unified Action Conditioning for World 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
Bowen Xue, Zizhang Li, Hong-Xing Yu, Jiajun Wu
UniAction represents each action with semantic identity and video-space/time coordinates, then processes action and video tokens together through a pretrained video model with calibrated attention. It is tested on force, camera, hand-pose, and robot-pose control using four datasets. On force Action Following, it scores 0.7630 versus 0.5436 for a force-specific expert, in this tested setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.