Molmo2 ER
Molmo2 ER is a 4B model for embodied visual reasoning.
What the release provides
It adapts Molmo2 4B with scene understanding, spatial pointing, view correspondence and temporal reasoning. It supplies the vision-language backbone for MolmoAct2. Its training uses specialization followed by joint refinement. Apache 2.0 applies.Molmo2 ER official model card
How to use it well
Practical guidance. Evaluate each spatial skill needed by the application. A correct scene description does not establish that a generated action will work on a physical robot.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | allenai/Molmo2-ER |