AI BriefingAllenAI unveils MolmoAct2
Key point
AllenAI has released the MolmoAct2 robot control model along with fine-tuned versions.
Details
MolmoAct2 is a 5B-scale vision-language-action (VLA) robot control model released by Ai2 (AllenAI). It attaches robot state and action modeling to a Molmo2-ER-based backbone, targeting real-world deployment.
The official repository and Hugging Face include the base checkpoint along with fine-tuned versions MolmoAct2-LIBERO, MolmoAct2-DROID, MolmoAct2-BimanualYAM, and MolmoAct2-SO100_101.
LIBERO: a general robotics version for the Spatial/Object/Goal/Long mixed benchmarkDROID: absolute joint-pose control based on filtered DROID Franka mixed dataBimanualYAM: based on 700 hours of bimanual manipulation demonstrations, among the largest public bimanual datasetsSO100_101: absolute joint-pose control based on the SO-100/SO-101 dataset
The official materials also state that model weights, training code, and the entire training dataset have been released, and the model is provided under the Apache 2.0 license.
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
Stanford Researchers Propose EXPO-FT for Universal Post-Training in Robotics
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance Metrics