China's first five humanoid industry standards grade robots L0 to L5 and set rules for training data
MIIT published five humanoid robot industry standards, SJ/T 12353 to 12357, which took effect on release on Nov. 16. They cover intelligence grading, joints, dexterous hands, training grounds and training data: how robots learn and how buyers rank them, ahead of any standard on safety beside workers.
China's industry ministry has published its first five industry standards for humanoid robots. Numbered SJ/T 12353 to SJ/T 12357, they were listed in MIIT Announcement No. 30 of 2026, released on Nov. 16, and took effect on release. Only 20 of the 274 standards in that announcement did; the other 254 start between February 2027 and November 2028. The five come from MIIT's humanoid robot and embodied AI standards committee, TC8, The Paper reported.
The first batch is written for buyers and data collection more than for the shop floor. The headline standard grades humanoid intelligence from L0 to L5 across four dimensions: perception and cognition, decision-making and learning, execution, and collaboration and interaction. The ministry's list says it is meant for classifying intelligence "in R&D, testing, evaluation and application selection", which gives a buyer a level to ask for. Two of the other four are about how robots learn. The training-ground standard covers zoning, layout, equipment, safety and ethics, environment control and networks; the training-data standard covers the whole lifecycle, from collection to transfer, use and destruction. No standard in the batch is about safety beside workers, although the joint standard includes safety requirements.
Hands get their own standard, and joints leave them out
The joint standard covers motor-driven integrated joints, with requirements for structure, interfaces, performance, safety and electromagnetic compatibility. It explicitly excludes the small joints used in dexterous hands. Those fall under the dexterous hand standard, which covers appearance and degrees of freedom, motion, sensing, grasping and manipulating objects, interfaces, reliability and environmental adaptability, plus modeling and simulation. The Paper said it targets inconsistent interfaces and inefficient integration of hands into whole robots.
Newsletter
Robrief Weekly
One email a week with the robotics news that matters. Free, unsubscribe anytime.
Five of at least 100
MIIT's draft guide for a national humanoid standard system, opened for comment in August, calls for at least 100 key standards by 2028, covering capability testing and evaluation, key technologies, platforms and systems, applications and safety governance, Xinhua reported. Capability grading was on that list, and so were product identification codes, which MIIT has since turned into draft rules giving every robot an ID code.
Korea started at the other end. Its state robot agency is building a humanoid safety certification, and LG is helping write the test. China's batch tells a buyer how capable a humanoid is; the safety-governance standards from the same plan were not in it.
Sources Sources verified
- 1.The Paper, Nov 17, 2026: 工信部发布5项人形机器人标准,涉及智能化能力分级、灵巧手及训练数据等thepaper.cn
- 2.Ministry of Industry and Information Technology, Nov 16, 2026: 中华人民共和国工业和信息化部公告(2026年第30号)wap.miit.gov.cn
- 3.Ministry of Industry and Information Technology, Nov 16, 2026: 公告(2026年第30号)附件1:274项行业标准 (PDF)wap.miit.gov.cn
- 4.Xinhua, Aug 24, 2026: 国家人形机器人产业标准体系建设指南将公开征求意见news.cn
- #China
- #MIIT
- #standards
- #policy
- #intelligence grading
- #dexterous hands
- #training data
- #TC8
Share this briefing
Know someone who follows robotics? Send it their way.
Related briefings
More in HumanoidsJPMorgan sees 300,000 to 400,000 US-made humanoids by 2030, a forecast that leans on Tesla
JPMorgan expects U.S. humanoid robot production of 300,000 to 400,000 units by 2030 and about 1.5 million by 2035. By the bank's own count of what other U.S. makers are aiming for, most of that volume has to come from Tesla.
Beijing's humanoid plan asks for robots that run 2,000 hours between failures, and capacity for 200,000 a year
Beijing's 2026–2028 embodied AI plan targets at least five world-class humanoid makers and capacity for 200,000 robots a year in the city by 2028. Its reliability targets say more: 2,000 hours between failures is about a year of single shifts.
GGII cuts its China humanoid forecast to 50,000-plus units and sees 15% of makers gone within three years
Gaogong Consulting now expects China to ship more than 50,000 humanoids this year, below the 60,000-plus it forecast in January, and says more than 15% of humanoid companies will fold or be bought within three years. Its price forecasts show where the squeeze comes from.
Newsletter
Robrief Weekly
One email a week with the robotics news that matters. Free, unsubscribe anytime.