Galaxea Open-World Dataset — Mobile Manipulation

Galaxea Open-World Dataset is a large-scale real-world mobile manipulation dataset containing 500+ hours of dual-arm robot data collected across residential, kitchen, retail, and office environments with fine-grained language annotations. Released in April 2026 under CC BY-NC-SA 4.0, the dataset is collected using the Galaxea R1 — a wheeled dual-arm mobile robot — performing whole-body manipulation tasks that require coordinating base movement with arm actions. The open-world collection methodology across four environment types (residential, kitchen, retail, office) provides the broadest environmental coverage of any Chinese-origin mobile manipulation dataset. Language annotations describe task goals, object properties, and action sequences at fine-grained resolution.

Dataset specifications
Year2026
Total hours500
EmbodimentsGalaxea R1 dual-arm mobile robot
Modalitiesrgb, depth, proprioception, language
Task categoriesmanipulation, pick-and-place, cleaning, cooking, long-horizon, human-robot-interaction
Data formatmp4, parquet, json
LicenseCC BY-NC-SA 4.0
Accessopen
MaintainerGalaxea Robotics
Origin countryCN

What is it?

Galaxea Open-World Dataset is a large-scale real-world mobile manipulation dataset containing 500+ hours of dual-arm robot data collected across residential, kitchen, retail, and office environments with fine-grained language annotations. Released in April 2026 under CC BY-NC-SA 4.0, it is collected using the Galaxea R1 — a wheeled dual-arm mobile robot — performing whole-body manipulation tasks requiring coordinated base movement with arm actions. The open-world methodology across four environment types provides the broadest environmental coverage of any Chinese-origin mobile manipulation dataset.

Who is it for?

Researchers working on mobile manipulation — the challenge of combining navigation with dexterous object interaction. Most manipulation datasets use stationary arm platforms; Galaxea Open-World provides one of the few large-scale datasets where the robot must move through the environment while manipulating objects. Language annotations make it suitable for language-conditioned policy training.

Key specifications

How it compares

Galaxea Open-World is the largest mobile manipulation dataset from a Chinese manufacturer. The Galaxea R1 platform is comparable to the Hello Robot Stretch (US) but with dual arms. AgiBot-Alpha covers more hours (1,979h) but uses a different robot platform without mobile base navigation. The four-environment diversity distinguishes it from single-setting mobile manipulation datasets.

Limitations and access notes

CC BY-NC-SA 4.0 prohibits commercial use and requires share-alike on derivatives. Chinese origin — geopolitical flag applies.

Linked professions

Frequently asked questions

What is the Galaxea R1 robot?

The Galaxea R1 is a wheeled dual-arm mobile robot developed by Galaxea Robotics. It combines a mobile wheeled base with two manipulation arms, enabling whole-body coordination between navigation and object manipulation — a key capability for service robots that must operate across entire rooms rather than at fixed workstations.

Can Galaxea Open-World be used commercially?

No. CC BY-NC-SA 4.0 restricts to non-commercial research and requires derivative datasets to share under the same license.

What environments does Galaxea Open-World cover?

Galaxea Open-World covers residential homes, kitchens, retail stores, and office spaces — four distinct environment categories requiring different navigation and manipulation strategies. This multi-environment approach is designed to force policies to generalise beyond a single familiar setting.

How do I access Galaxea Open-World?

Available on Hugging Face at huggingface.co/datasets/Galaxea/Galaxea-Open-World. No registration required.

What makes mobile manipulation harder than stationary arm manipulation?

Mobile manipulation requires coordinating base movement with arm manipulation simultaneously — the robot must navigate to the right position, maintain balance during arm motion, and handle the kinematic coupling between base and arm. Most manipulation policies assume a fixed base; mobile manipulation requires whole-body control that stationary datasets cannot train.