The Chinese Data Administration announced the preparation of data standards for embodied artificial intelligence. It is about AI systems operating by physical devices: humanoid robots, mobile robots and machines equipped with manipulators. The communication published on 13 September after gathering with manufacture representatives does not yet introduce a fresh standard. However, it shows that the office liable for data policy is active in the work on robotics.
The gathering was held on 10 September under the leadership of Liu Liehong, manager of the Chinese Data Administration. The conversation was attended by representatives of investigation institutes, technological companies and robotic training centres, including the Institute of Automation of the Chinese Academy of Sciences, Beijing Institute for General Artificial Intelligence, JD Group, the Humanoid Robot Data Center in the capital territory of Shijingshan and Beijing Humanoid Robot Innovation Center.
The administration reported that the fabricated AI sector is data-based and that the request for advanced quality, diverse and large-scale data is constantly increasing. It announced the improvement of data standards, coordination of the activities of local data authorities and support for enterprises that increase data spending.
Robot needs data on the physical world
The embodiment AI is not taught exclusively on texts, pictures or recordings. The robot control model must combine reflection of the environment with movement of the device and measure the consequence of the operation. erstwhile the device is to lift the cup from the table, it needs data about its position and shape, but besides about the position of its own arm, the gripping force on whether the object has slipped out during movement.
The data set for specified a task may combine images from cameras, depth information, strength sensor readings, joint positions, motion trajectory, command given to the robot and task consequence indication. This data must be recorded at the same time, decently described and verified. A synchronisation mistake between the image and the movement of the arm may teach a reaction model incompatible with the actual situation.
Therefore, in working on standards, it is not only the number of hours of exercise, the number of repetitions that counts. Common requirements may relate to the format of data, metadata, description of the scenario, the way in which activities are identified and quality control procedures. They besides facilitate the assessment of whether the 2 collections collected by different teams are suitable for akin applications.
Robot standards are already being developed
The data is part of a wider Chinese robotics standardisation system. In February presented a strategy of standards for humanoid robots and embodiment intelligence, comprising six areas:
- common grounds,
- intelligent calculations,
- components and limbs,
- complete robotic systems,
- applications and
- safety.
This structure besides includes issues related to the data cycle and model training.
As of 1 June, the manufacture standard YD/T 6770–2026 on the AI embodiment investigating methods applies. It describes how to prepare the test environment, build task libraries, conduct tests and calculate indicators in both simulation and real environment. This allows manufacturers and laboratories to measurement the operation of systems according to the same methodology.
Another document, YD/T 6771–2026, was approved on 24 July and will enter into force on 1 November. It refers to the quality of data sets for the embodiment AI. It introduces 8 evaluation areas:
- completeness,
- consistency,
- diversity,
- authenticity,
- ease of use,
- practical usefulness,
- expansion and
- safety.
It shall besides include rules on the organisation and verification of the data creation process.
From center to data training for the full industry
In June, the administration adopted a plan to build advanced quality manufacture data collections. According to this plan, by the end of 2028, a set of precedence areas, proven in applications, as well as tools and standards related to their production, is expected. The personified AI was mentioned alongside fields specified as intelligent driving, low-level economy and smart maritime economy.
The plan points straight to data from physical interactions, perception of environment and motion control. It besides allows the usage of simulations and synthetic data, which can grow the amount of examples needed during training. However, this does not replace the recording of device behaviour in a real environment where the robot meets different surfaces, weights of objects, lighting and human movement.
Such data is collected in specialised centres. An example is described earlier by us Beijing Humanoid Robot Innovation Centerwhere robots execute tasks modeling work at home, store, office, mill and care facility. Movements, sensor information and task consequence are recorded for each action.
It is at the interface of specified centres, model companies and data management authorities that there is simply a request for common rules for assessing and describing training material.
Sources
- Chinese Data Administration, "Liu Liehong presided over a symposium on embodiment intelligence", 13 September 2026: https://www.nda.gov.cn/sjjj/jgsz/jld/llh/llhldhd/0913/2026091310473055195636_pc.html
- Chinese Data Administration, "Action Plan for building high-quality manufacture data sets", 8 June 2026: https://www.nda.gov.cn/sjj/zgk/zcfb/0608/20260608172117399715004_pc.html
- Xinhua, ‘Standards for embodiment intelligence are being developed’, 1 June 2026: https://www.news.cn/20260601/67740ab4f48245679503782b8dce8a05/c.html
- Digital China Summit, "First Quality Standard of Data Collections for Embossed Intelligence", 4 August 2026: https://www.szzg.gov.cn/2026/xwzx/szkx/202608/t20260804_5354035.htm
- Pandaily, "National Data Administration to advance Embodied-Intelligence Data Standards in Due Course", 14 September 2026: https://pandaily.com/national-data-administration-embodied-ai-data-standards
Email: [email protected]
![Ostrołęka ma dwóch nowych prokuratorów. Nominacje odebrali z rąk Żurka i Korneluka [ZDJĘCIA]](https://www.eostroleka.pl/luba/dane/pliki/zdjecia/2026/fad63e92-3c7d-48a8-8cc8-8b4259c570e4.jpeg)