AI BriefingAllen Institute Unveils MolmoMotion
Key point
A model and dataset called MolmoMotion have been released that predict the future 3D trajectory of objects based on language instructions.
Details
Allen Institute has announced MolmoMotion, which predicts the future 3D point trajectory of objects based on text instructions and video frames. Instead of relying on conventional rendering approaches, this model uses 3D points in world space to improve computational efficiency, and it has a Class-agnostic property that allows it to be applied regardless of object type.
The key features and released assets are as follows:
- MolmoMotion model: Uses Molmo 2 as its backbone to connect language instructions with objects/points in images, and can be used directly as input for robotics planning and video generation models.
- MolmoMotion-1M: The largest 3D point trajectory dataset with action descriptions, extracted from 1.16 million videos.
- PointMotionBench: A benchmark based on 2,700 video clips designed to measure the accuracy of object-centric 3D motion prediction.
This model can be utilized for various downstream tasks such as robotics planning and trajectory-conditioned video generation.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
OpenComputer Hosts 'Ship Video' Agent Using Claude Opus 5.5 to Generate Explainer Videos from Code
Stanford Researchers Propose EXPO-FT for Universal Post-Training in Robotics