Data supply

Embodied AI training data supply: training grounds, gyms and public datasets

Who is collecting the hours? The ledger holds 15 data supply records as of September 2026 across South Korea, Japan, United States, Global, Germany, European Union, China: 9 verified at a primary source, 0 announced with no document yet, 1 searched and absent, 0 open. Every row links the document it rests on and the date it was checked.

China

6 records

VerifiedData supplyEAD-2026-0013
The largest open real robot dataset is Chinese, over a million trajectories, and licensed for non commercial use only
China · verified 13 September 2026
GitHub release notes: Alpha, 92,214 trajectories, 30 December 2024; Beta, 1,003,672 trajectories, about 43.8 terabytes, 1 March 2025.
Reported, primary not reachedData supplyEAD-2026-0010
Shanghai's Zhangjiang humanoid training ground opened its first phase with over 100 heterogeneous robots
China · verified 13 September 2026
February 2025, Liberation Daily on the Shanghai portal: first phase deployed over 100 heterogeneous robots.
Reported, primary not reachedData supplyEAD-2026-0009
Beijing's Shijingshan humanoid data training center reports close to 270 robots producing about 20 million data items a year
China · verified 13 September 2026
May 2026, Beijing Daily on the municipal portal: close to 270 robot units training at full load, annual data output close to 20 million items.
Reported, primary not reachedData supplyEAD-2026-0008
Beijing's humanoid robot innovation center runs a data and training base with over 120 robots in more than 30 scenes
China · verified 13 September 2026
Reported in March 2026: close to 5,000 square meters, more than 30 typical scenes, over 120 mainstream robot units, tens of thousands of hours delivered.
Reported, primary not reachedData supplyEAD-2026-0004
The hours gap: a CAICT report puts the need on the order of ten million hours and the world's usable supply at under a million
China · verified 13 September 2026
The report's framing, per the Xinhua summary: a 55 billion parameter model needs about 1.1 trillion tokens; if all of it came from real robot data, that is roughly 21,200 robots working a full year, or effective data at the ten million hour scale.
Reported, primary not reachedData supplyEAD-2026-0003
China has over 70 embodied AI training grounds in use, per a CAICT report the ledger could only reach through state media
China · verified 13 September 2026
CCTV, reproduced on a government portal: over 70 training grounds built and in use nationwide, and forty odd under construction or planned. The Xinhua summary carried by IT Home gives the second figure as 46.

European Union

1 record

AbsentData supplyEAD-2026-0016
No EU or member state program funds robot training data collection as its object
European Union · verified 13 September 2026
euROBIN, coordinated by DLR, runs from July 2022 to December 2026 with an EU contribution of 11,499,999 euro; its stated exchange is software, data and knowledge over the EuroCore repository among its own labs.

United States

3 records

VerifiedData supplyEAD-2026-0034
NVIDIA's synthetic supply: 780,000 trajectories in eleven hours, and an open simulated set of about 273,000
United States · verified 13 September 2026
NVIDIA newsroom, 18 March 2025: 780,000 synthetic trajectories generated in eleven hours; the developer blog of the same day says over 750,000.
VerifiedData supplyEAD-2026-0033
DROID: 76,000 demonstration trajectories, 350 hours, one robot arm, under CC BY 4.0
United States · verified 13 September 2026
Project page: 76,000 demonstration trajectories or 350 hours of interaction data, 564 scenes, 86 tasks; the paper abstract gives 84 tasks.
VerifiedData supplyEAD-2026-0029
The US answer to a training ground is a manufacturing institute's data call: the ARM Institute's AI Data Foundry
United States · verified 13 September 2026
ARM Institute, 5 March 2025: a key barrier to the adoption of AI within robotics for manufacturing is a lack of real, quality data to train AI models.

Germany

1 record

VerifiedData supplyEAD-2026-0020
Europe's nearest equivalent to a training ground is one company: NEURA Robotics has ten gyms under development
Germany · verified 13 September 2026
NEURA, 22 July 2026: ten NEURA Gyms currently under development; five locations across Europe, the United States and China named as the year end target.

Japan

1 record

VerifiedData supplyEAD-2026-0035
Japan funds a robotics data platform and a physical AI foundation model program through NEDO, with no published data volume
Japan · verified 13 September 2026
AIRoA, 26 September 2025: selected by NEDO for development of data platforms for generative AI foundation models in the robotics field, 1 October 2025 to 31 August 2029, covering data collection, model development and validation.

South Korea

1 record

VerifiedData supplyEAD-2026-0036
South Korea's public robot dataset: 209,403 grasp videos that only Korean nationals may request, for non commercial use
South Korea · verified 13 September 2026
AI Hub dataset 71713, robot behaviour data for small object grasping: about 209,403 videos, version 1.2 released 4 December 2024; the page states that only domestic nationals may request the data.

Global

2 records

VerifiedData supplyEAD-2026-0024
The largest open supply of hours is human, not robot: Ego4D holds over 3,670 hours of first person video
Global · verified 13 September 2026
Ego4D site: over 3,670 hours of daily life activity video, 923 unique participants, 74 locations in 9 countries, released 17 February 2022.
VerifiedData supplyEAD-2026-0023
Open X-Embodiment: over a million real robot trajectories from 22 embodiments, pooled from 60 datasets, under CC BY 4.0
Global · verified 13 September 2026
Project page: over one million real robot trajectories, 22 robot embodiments, 60 existing datasets from 34 robotic research labs.

Every surface

Cut the ledger another way

By kind

By jurisdiction

Cite this page

Free to reuse under CC BY 4.0, with attribution.

In a sentence: According to the GAGE Embodied AI Data Ledger (as of 13 September 2026), embodied ai training data supply: training grounds, gyms and public datasets.
APA: GAGE (Global Academy of Generative-AI Education). (2026). Embodied AI training data supply: training grounds, gyms and public datasets. Embodied AI Data Ledger. Retrieved 13 September 2026, from https://www.gage.academy/tools/embodied-ai-data-ledger/kinds/supply
MLA: "Embodied AI training data supply: training grounds, gyms and public datasets." Embodied AI Data Ledger, GAGE (Global Academy of Generative-AI Education), 13 September 2026, https://www.gage.academy/tools/embodied-ai-data-ledger/kinds/supply.
Chicago: GAGE (Global Academy of Generative-AI Education). "Embodied AI training data supply: training grounds, gyms and public datasets." Embodied AI Data Ledger. Last modified 13 September 2026. https://www.gage.academy/tools/embodied-ai-data-ledger/kinds/supply.
Permalink: https://www.gage.academy/tools/embodied-ai-data-ledger/kinds/supply

Last updated 13 September 2026. Every record re verified 13 September 2026. The ledger is checked every Monday, and the same day for any announcement by a data regulator.

Back to the full ledger.

GAGE briefings tell you which AI regulation deadlines are coming, what they actually require of you, and when a program opens.