Skip to content

Dataset · allenai

Data mixture for robotics model pre-training

allenai/MolmoAct-Pretraining-Mixture

A multimodal corpus of trajectories and action reasoning for training robotic manipulator control models.

What's inside

This dataset from the Allen Institute for AI contains between 10 and 100 million image-text pair examples used to pre-train the open MolmoAct model. It includes the OXE subset reformatted into action reasoning data, auxiliary robot data, and links to multimodal web data. It also features a set of 10,000 high-quality trajectories from a single-armed Franka robot performing 93 unique tasks.

What you can do with it

  • Pre-training vision-language-action (VLA) models for robotic manipulator control
  • Training systems for physical action planning and reasoning based on visual data
  • Fine-tuning robotic manipulation models on Franka robot trajectories
Tasks
Robotics
Size
10M–100M rows
Downloads this month
19.8K
Likes
14

How to load it

from datasets import load_dataset

ds = load_dataset("allenai/MolmoAct-Pretraining-Mixture")