Reproduce “MolmoMotion: Forecasting Point Trajectories in 3D with Language Instruction” 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

Jianing Zhang, Chenhao Zheng, Yajun Yang, Max Argus, Rustin Soraki, Winson Han, Taira Anderson, Chun-Liang Li and 5 more

MolmoMotion forecasts future metric 3D positions of selected object-attached points from short RGB histories and action instructions. It uses a Molmo2 backbone and anchor-relative coordinates, with autoregressive and flow-matching variants. On HOT3D, the three-frame autoregressive model reaches 0.109 m ADE versus 0.129 m for ObjectForesight. This comparison is limited to HOT3D and differs in text conditioning and mesh requirements.

Summary by alphaXiv.

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