EmbodiedHub LogoEmbodiedHub
Back to Search
O

Open X-Embodiment

DatasetActive

Open X-Embodiment is a large-scale collaborative open-source dataset led by Google DeepMind with 34 research institutions worldwide. Contains over 1 million episodes across 22 robot embodiments and 500+ skills, standardized in RLDS format. Before Open X-Embodiment, every lab collected data in incompatible formats. This dataset unified robot data across platforms, enabling cross-embodiment training for the first time. Used to train RT-1-X, RT-2-X, OpenVLA, Octo, and pi0. Analogous to ImageNet in computer vision or Common Crawl in LLM training — the canonical dataset that made generalist robot policies feasible.

Details

Updated:6/20/2026
https://robotics-transformer-x.github.io
sample count1000000
licenseCreative Commons Attribution 4.0
modalityvision, proprioception, actions

Tags

multi-embodimentVLA-trainingcross-embodimentopen-sourceRLDSGoogle-DeepMindlarge-scalerobot-manipulation

Relationships

Google DeepMind
Company
DEVELOPED_BY
RT-2-X
Model
Used to Train
RT-1-X
Model
Used to Train
RT-X
Model
Used to Train
Octo
Model
Used to Train
RT-2
Model
Used to Train
OpenVLA
Model
Used to Train
π0 (pi-zero)
Model
Used to Train
RoboNet
Dataset
SUBSET_OF
BC-Z
Dataset
SUBSET_OF
Open X-Embodiment
Dataset
RELATED_TO
LeRobot
Project
USED_BY

Sources

https://robotics-transformer-x.github.io
website
Visit
https://arxiv.org/abs/2212.06817
paper
Visit

Appears In

Embodied AI Landscape

An entry-point knowledge page providing an overview of the embodied AI ecosystem, including companies, robots, models, datasets, and open-source tools.

Embodied AI Datasets

A collection of key datasets used for training embodied AI models, including robot manipulation and vision-language-action models.