MolmoAct2 runs zero-shot on an SO-ARM101, no training required 🦾
@allen_ai open Action Reasoning Model is now in LeRobot with the full lifecycle covered: fine-tuning (full or LoRA), evaluation, and real-robot deployment. It pairs a Molmo2-ER vision-language backbone with a flow-matching action expert to turn images, language, and proprioceptive state into action chunks.
The best part: ready-made checkpoints ship with calibration correction baked in, so you can point it at your SO-100/101 and just watch it work. Inference fits in ~12GB at bf16, and LoRA fine-tuning runs on a single 24GB GPU.
Big thanks to the Ai2 team for building this in the open.
about to hop on my 17 hour flight back home 🇸🇬; SF you've been amazing 🫶 until next time!!!!!!!!!!!
i think i'll make time and work hard to come back for @vercel ship in october!!!!!!!!! RAHHHH
Open science robotics foundation models are the fastest way to get robots into the real world.
MolmoAct 2 has now seen 600k+ downloads across its models and datasets.
Cortex AI contributed high-quality, long-horizon real-world data for the bimanual YAM setup and conducted large-scale real-world evaluations across multiple policies.
We'll keep pushing open science to bring robot intelligence out of the lab and into the real world.
Stay tuned.