Reproduce “Inertial Video 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

Yuan Gao, Fan Xu, Yu Guo, Yuxu Lu, Zhengru Fang, Hao Wu, Xiaomeng Huang

InertialWorld models action-conditioned video by carrying scene evidence forward as an evolving state. It combines a scene-aligned dense grid with learned regional carriers that move with local content, predicts a complete next frame, and gates updates to memory. On RT-1, 32-frame episode-averaged PSNR was 24.37 dB versus 21.95 dB for RLVR-World; this checkpoint comparison is not clean held-out generalization.

Summary by alphaXiv.

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