The humanoid robot industry is finally getting its own "national standards." The Ministry of Industry and Information Technology recently released the draft for public comment on the National Humanoid Robot Industry Standard System Construction Guidelines (2026 Edition), with the official public comment period running from August 25 to September 23. This document may appear to be a mere set of "technical specifications," but it carries profound implications: it represents China's first systematic framework for a humanoid robot standard system. Rather than patching up individual technical indicators, it establishes the industry's "weights and measures" from the top down, starting with the fundamental design.

Governmental policy momentum for the robotics sector is now undergoing a critical transition—shifting from the first half, which focused on "providing funding and preferential policies," to the second half centered on "defining standards, opening up application scenarios, and shaping the rules of the game." According to the draft guidelines, by 2028, China aims to finalize at least 100 key standards covering areas such as humanoid robot capability testing and evaluation methods, core technologies, platforms and systems, scenario applications, and safety governance. Furthermore, the initiative seeks to promote the implementation of these standards across more than 200 enterprises. This signals an important shift in the policy logic for China's humanoid robot industry: previously, the focus was on "whether to develop robots," but now it is on "how robots can truly form a fully-fledged industry."

Establishing the "Ruler" First

So why do humanoid robots need a more comprehensive standard system precisely at this moment? By 2025, the number of domestic humanoid robot integrator companies had already surpassed 140, with over 330 products launched. By the first half of 2026, the MIIT further disclosed that more than 230 enterprises had released over 400 complete robot products, with annual production projected to exceed 100,000 units. As product variety and technical pathways proliferate, the industry increasingly requires a common "ruler" for measurement. One company might claim its robot is "highly intelligent," while another emphasizes "high dynamics"; some compete on speed, others on payload, joint degrees of freedom, or battery life. Without unified capability grading, interface specifications, and testing methods, many parameters are difficult to compare on a level playing field.

When robots transition from the laboratory and exhibition halls to factories, this problem is further amplified: picking up a cardboard box at a fixed location is a completely different difficulty level from continuously processing parts of varying shapes on an assembly line. Similarly, walking 100 meters on a flat surface is not the same as working continuously for hours in mining, power inspection, or warehousing environments. Therefore, the essence of standardization is not to "limit" innovation but to create a common language for the industry. This draft guidelines proposes establishing a sound humanoid robot standard system, focusing on breakthroughs in six major areas: fundamental commonalities, brain-inspired and intelligent computing, limbs and components, complete machines and systems, applications, and safety ethics. It also clarifies directions for capability grading, unique identification, industrial scenario operational capability, environmental adaptability, and ethical and social impact assessments.

The "unique identification" aspect is particularly noteworthy. In the future, a robot may not just be a hardware product; it will need identity information spanning the entire lifecycle of production, deployment, operation, and maintenance: who manufactured it, what components were installed, how long it has been running, what faults have occurred, and what tasks it has performed. This information could all be integrated into a more standardized lifecycle management system.

However, for a "ruler" to be meaningful, it must be tested somewhere. Robots don't just need prettier data from laboratory settings; they need results derived from real factories, real warehouses, and real power grids. This is where the second lever of policy comes into play, targeting the places that truly determine robot value: real-world scenarios.

Pushing Robots into "Real Environments"

If standards address "how to judge if a robot is good," then scenarios address another crucial question: is the robot actually useful? Policy changes have already become quite apparent. The 2026 Government Work Report continued to propose cultivating the development of embodied intelligence, further clarifying the encouragement for central and state-owned enterprises to take the lead in opening up application scenarios, while promoting the commercialization and large-scale application of AI in key industries. The shift from "cultivation" to "opening scenarios" reflects a change in policy KPIs. The biggest contradiction in the robotics industry today is no longer just "whether robots exist," but that a vast number of robots still lack replicable job roles that can be sustained over time.

Why is it essential to enter "real environments"? Because embodied intelligence has a natural characteristic: the deeper robots go into the real world, the more the real world trains them in return. Only by being in the field can robots continuously accumulate real-world data—full-body motion trajectories, force-position control curves, operation sequences, anomaly states, and edge conditions—which is then used to train their models. More importantly, this round of policy shifts the evaluation focus of robots from "demonstration effects" directly to "commercial value." The special action plan explicitly proposes to scientifically assess the real task success rate, efficiency improvement rate, safety reliability, and economic feasibility of complete robot solutions. These indicators mean that evaluating a robot will no longer be just about "can it do it," but also: how long can it do it continuously? How many times out of 100 tasks will it succeed? How much efficiency does it improve compared to manual labor? What's the probability of an accident? And ultimately, does the cost-benefit equation work out?

"National Team" Isn't Just Buying Robots

Who is best suited to provide these real-world scenarios? One answer is the central state-owned enterprises (SOEs). Many of the scenarios China's next-stage robots need to enter—power, energy, mining, metallurgy, ports, logistics, space, and deep sea—are precisely concentrated in the hands of large SOEs. In July 2026, the State-owned Assets Supervision and Administration Commission of the State Council launched the "Scene Vientiane" special action for central enterprises, requiring the establishment of open scenario lists and the creation of demonstration benchmarks. It also released the first batch of 10 landmark open scenarios, which include "embodied intelligence empowering intelligent manufacturing of coal machinery equipment."

Previously, the industrialization path for robot companies often followed this pattern: first build a machine, then find customers everywhere. The difficulty with this model is that truly high-value industrial scenarios are hard to access. Companies might not have long-term opportunities to enter power grids, mines, or ports, let alone obtain sufficiently complex operational data. Robots can't train for extended periods, and users hesitate to purchase at scale because the products haven't been validated—creating a "chicken-and-egg" cycle. Now, policy is attempting to break this cycle. The real-scenario training special action plan explicitly proposes forming "innovation application consortiums" around each scenario: user units are responsible for opening up real training spaces and quantifying deployment targets; integrator companies handle robot capability and scenario adaptation; and model algorithm companies, component suppliers, and research institutions collaborate on tackling key challenges. The industry organization model is moving from "a single robot company fighting alone" to "scenario owners + integrators + model companies + component suppliers + research institutions" completing a real task together.

The Next Round of Elimination Has Already Begun

With standards in place, training grounds being built, and the national team opening up scenarios, does this mean anyone making humanoid robots can reap policy dividends? Quite the opposite. The more complete the standards and the more realistic the scenarios, the faster industry shakeout may occur.

The first change is a shift from "parameter competition" to "comparable competition." In the past, different manufacturers used their strongest individual metrics to tell stories, making it hard for the market to truly judge product quality. As capability grading, complete machine testing, interface standards, safety protocols, and scenario evaluation systems are established, robots will increasingly be compared under the same set of rules, much like cars and industrial equipment. Who is stable, who is safe, and who has a high task success rate will become increasingly transparent.

The second change is moving from "selling machines" to "selling productivity." The real-scenario training special action plan explicitly encourages exploring a "humanoid robot as a service" model, lowering user entry barriers through pay-for-performance and operating leases. It also proposes coordinating equity, debt, and insurance tools, while exploring dedicated robot insurance. In the future, customers may care less about whether a robot costs 500,000 yuan or 300,000 yuan, and more about: how much production efficiency can I get back for the money I spend? Robot companies' revenue structures may also shift from one-time hardware sales to a combination of complete machines, software, operations and maintenance, industry solutions, and RaaS (Robot-as-a-Service).

The third change is the redistribution of industry chain value. Once standards and large-scale applications are established, the beneficiaries may not be limited to integrator companies. Core components, dexterous hands, sensors, testing and certification services, training grounds, data collection, simulation platforms, robot safety, operations and maintenance services, and even insurance could all gradually grow into new industry segments. Those who can enter real scenarios first and continuously generate data will be more likely to accumulate industry know-how, customer stickiness, and standard-setting influence.

The final change may be the harshest: policy is beginning to press the accelerator and the brake simultaneously. Officials from the MIIT have previously advised local governments to avoid blindly following trends and rushing into action, recommending they choose specific niche tracks based on resource endowments, industrial foundations, and research strengths, rather than pursuing a "big and comprehensive" approach. Enterprises are also advised to identify their comparative advantages and not chase trends indiscriminately. This means policy does not encourage "every region building a robot industrial park" or "every company making humanoid robots," but rather robots that can pass standards, enter scenarios, solve real problems, and ultimately achieve a viable business model.

From this moment on, the real competition in the robotics industry is no longer about who looks the most like the future, but who can become reality first. In 2023, China drew a line for a future industry track for humanoid robots. By 2026, policy is paving that track, drawing traffic lines, setting the rules of the road, and inviting the first batch of real users to take the wheel.