MolmoAct Dataset: Robot Manipulation Demonstrations (Ai2)

Provider
Allen Institute for AI (Ai2)
License
CC-BY-4.0 (source)
Size
7,912 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/allenai/MolmoAct-Dataset)
Access
Free and public. No account with the data provider is needed.
Category
Robotics
Source
https://arxiv.org/abs/2508.07917

About MolmoAct Dataset: Robot Manipulation Demonstrations (Ai2)

The MolmoAct Dataset holds the robot demonstrations Ai2 collected in house to train MolmoAct, an action reasoning model that plans in space before acting (Lee et al., arXiv:2508.07917). It is split into two LeRobot-format subsets, molmoact_dataset_household and molmoact_dataset_tabletop, each with a train split of Parquet data and metadata in the standard LeRobot layout. Measured 2026-11-14, the repository holds about 814 GB. Episodes pair camera observations with robot state, actions and language instructions for household and tabletop manipulation tasks. Load either subset with lerobot's LeRobotDataset by pointing at its folder, or read the Parquet files with pandas or pyarrow. Useful for training and evaluating vision-language-action models, spatial reasoning for manipulation and imitation learning. The card states the dataset is licensed under CC BY 4.0 and asks that use follow Ai2's Responsible Use Guidelines; cite the MolmoAct paper.

How to load MolmoAct Dataset: Robot Manipulation Demonstrations (Ai2) in Python

from datasets import load_dataset
ds = load_dataset("allenai/MolmoAct-Dataset")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: molmoact_dataset_household/train/data/chunk-000/episode_000000.parquet

Analyze this dataset without downloading it

On Scigantic, MolmoAct Dataset: Robot Manipulation Demonstrations (Ai2) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Robotics datasets