On Monday 24 August 2026, Xinhua reported from Beijing that China’s Ministry of Industry and Information Technology had put out a draft “National Humanoid Robot Industry Standard System Construction Guide (2026 edition).” No new robot walked out of that filing. The useful output is a worklist: at least 100 key standards by 2028, covering capability testing and evaluation, key technologies, platforms and systems, application scenarios, and safety governance, with more than 200 enterprises supposed to promote and implement them.
That is a standards audition. It is not a product launch, and it is not a spec sheet.
What Monday’s draft actually put on paper
Xinhua’s English copy, timestamped 25 August 2026 at 00:08 and datelined Beijing 24 August, defined a humanoid as a new generation of artificial-intelligence-driven intelligent terminals with human-like forms and capabilities that can autonomously perform complex tasks. The Chinese Xinhua file, posted the same evening at 21:27, used the same definition. Both wires said rapid development of the sector has created an urgent need for common standards covering terminology, interfaces, testing, and evaluation, to guide technological innovation, product development, and large-scale application. More than 100 humanoid-robot-related standards, Xinhua English wrote, are already under development or in preliminary research in China and abroad. A construction guide is how you organise that pile. It is not how you ship a knee.
Public comment runs from Tuesday 25 August through 23 September 2026. MIIT’s Department of Science and Technology, in a notice dated 24 August and carried on 25 August by Gkzhan, asked anyone with comments to fill in a feedback form and send it to kjbz@miit.gov.cn, with the subject line naming the guide. The notice lists the draft as Attachment 1 and the form as Attachment 2.
The Chinese Xinhua file named the six areas the draft wants to break through: foundational and common standards, brain-inspired intelligence and computing, limbs and components, complete machines and systems, applications, and safety and ethics. Digital Watch, writing the same day, put the same six headings on the framework. The construction principles, both Xinhua copies said, are to be systematic and comprehensive, to put urgent needs first, and to allow dynamic updates. By 2028 the draft aims to have a systematic, science-based, coordinated standard system, and to have more than 200 enterprises carrying out publicity and implementation. Those are planning sentences. They are not a published test method.
Digital Watch also described the framework as running from terminology, classification, digital identity, testing, reliability, and interoperability through data, models, operating systems, and human-machine interaction. I will take that as a map of the rooms, not as a list of finished documents. A room labelled “limbs and components” is still empty until someone writes the continuous-torque method that goes in it.
Unique identifiers, capability grades, and the factory floor
The parts that matter to anyone who buys a payload claim, a joint, or a thermal story sit in four lines of the draft.
First, capability grading. The draft calls for standardizing methodologies for grading humanoid-robot capabilities. That is the closest thing in the document to a common language for “this arm can hold five kilograms.” Today every brochure uses a different wrist, a different elbow angle, and a different time window. A peak newton-metre on a cold motor is not a shift-length number. A lab lift with the elbow tucked is not a tote held off-axis at arm’s length. If a grade actually lands, a 3 kg claim and a 20 kg claim stop being the same kind of sentence. Until then, capability is marketing copy.
Second, unique identifiers. The draft proposes standardizing the definition of unique identifiers, coding rules, and identification mechanisms for humanoid robots. Digital Watch put digital identity in with terminology and classification. A serialised body is how you keep a thermal map, a firmware revision, and a safety case attached to the same machine when it moves between a lab, a line, and a scrap pile. Without a unique ID, “the robot that overheated on Tuesday” is a story, not a record. With one, a buyer can finally ask which unit, which software build, and which duty cycle produced the claim on the card.
Third, industrial work. For structured and semi-structured industrial environments, the draft proposes requirements covering operational capabilities and environmental adaptability. Xinhua’s Chinese file specified manufacturing-process operation capability and environmental adaptability. Xinhua English added the practical complaint that anyone who has stood next to a cell already knows: reflections and dust can disrupt perception, and continuous operation places heavy demands on durability, heat management, and component life. Common interfaces and safety standards, that copy said, are emerging as practical issues in industrial deployment. Industrial production leaves little room for error. Those are not unveil sentences. Those are the sentences a shift supervisor already writes on a whiteboard.
Fourth, safety and ethics. Beyond the technical and industrial specifications, the draft calls for standards governing ethical requirements and social-impact assessments in the development and application of humanoid-robot technologies, placing safety and ethics alongside performance and commercialization. Digital Watch listed physical safety, cybersecurity and data protection, model and algorithm safety, behavioural safety, application safety, and ethical and social governance across the lifecycle. I will not turn that list into a finished safety case. It is a table of contents for work that has not been written yet.
What the draft still does not give a buyer
This is a construction guide, not a standard. A guide tells you which standards to write. It does not publish a torque table, a payload test fixture, a thermal soak, or a battery duty cycle. I will not invent one. I will also not treat the surrounding 2026 paperwork as if it were this draft. Xinhua English noted that in June, MIIT and China’s state assets regulator launched a training initiative aimed at application validation and regular deployment in representative scenarios by the end of 2026, and at deployment capacity at the 10,000-unit scale. At a 20 July press briefing, MIIT said China had developed more than 400 humanoid-robot models, over half of the global total. Digital Watch added that in June the ministry also sought comments on ten proposed industry-standard projects covering humanoid robots, including technical requirements for use in substations. Those are the neighbouring files. Monday’s filing is the system map.
A buyer still cannot read a common payload definition, a continuous joint-torque method, or a heat-soak that every OEM has to run. If unique identifiers and capability grades actually land, that is the first time a procurement sheet and a brochure might use the same words for payload, joint, and thermal claims. Until then, the 100-standard target is a worklist with a comment window. Comments close on 23 September 2026. After that, the draft can still change. Treat “national standard system” as an intention until the methods exist on paper and someone runs them on a hot motor.
FIRGELLI builds the linear actuators you put under a lift, a fixture, a hatch — the boring layer around a cell. Nobody in the MIIT draft announced one of those joints, and I will not pretend they did. What Monday announced is who has to write the test methods, and how little of the hardware those methods would measure is public while they do it.