AI Tool Profile
MolmoMotion
MolmoMotion forecasts language-conditioned 3D point trajectories from short visual histories and selected object points.
Verification & Sources
- Status
- Verified
- Source links
- 5
- Freshness
- Verified July 10, 2026
- Last verified
- July 10, 2026
- Last updated
- July 10, 2026
Key source checks
Suggest a correction
What It Does
MolmoMotion forecasts language-conditioned 3D point trajectories from short visual histories and selected object points.
Full Guide
MolmoMotion is an open vision-language model from Ai2 for forecasting how selected points on objects will move in three dimensions. It takes a short visual history, user-selected query points, and a natural-language action description, then predicts future 3D point trajectories.
What it does
Ai2 released autoregressive and flow-matching variants for different forecasting needs. The release also includes MolmoMotion-1M, a dataset of action-described 3D trajectories, and PointMotionBench, a human-validated benchmark. Ai2 describes robotics planning and trajectory-conditioned video generation as possible downstream research uses.
Access and limitations
The model weights, dataset, benchmark, and code are published through Ai2, Hugging Face, and GitHub. The model card describes MolmoMotion as a research and educational release and advises validating predicted trajectories before using them to drive an actuated system.
Who it is for
MolmoMotion is intended for researchers and developers working on 3D motion forecasting, robotics planning, vision-language systems, and controllable video research.
Official sources
Tool Links
Launch History
MolmoMotion
Ai2 released MolmoMotion models, code, MolmoMotion-1M data, and PointMotionBench for language-guided 3D motion forecasting.
- Kingy
- Needs review
- Demo
- Needs review
- YouTube
- Needs review
Ai2 released MolmoMotion as a complete open stack — models, code, the MolmoMotion-1M dataset, and PointMotionBench for language-guided 3D motion forecasting (allenai.org). AI Researchers,…