Ai2 (Allen Institute for AI)MolmoAct2
MolmoAct2
MolmoAct2 connects visual-language reasoning to a continuous robot action expert.
Research release
What the release provides
A flow-matching action expert receives per-layer key-value conditioning from the Molmo2 ER backbone. The multi-embodiment checkpoint is a fine-tuning foundation, not a ready-to-run policy for one robot.MolmoAct2 official model card
How to use it well
Practical guidance. Match action normalization, camera inputs and robot embodiment to the checkpoint. Validate predicted actions in the intended environment before allowing physical movement.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | allenai/MolmoAct2 |