MolmoAct

MolmoAct

Open multimodal language model for action

Description

Robot action models are mostly black boxes, hard to understand or correct. MolmoAct from Ai2 is an open multimodal action reasoning model that plans trajectories in space before acting, with visible, steerable reasoning.

The faster, stronger MolmoAct2 is out, with models, data and training code, deployable zero-shot on arms like Franka.

Features



Spatial reasoning:Plans first.

Explainable:Visible process.

Steerable:Adjust trajectories.

Fully open:Models and data.