China Publishes First Six Humanoid Robot Standards

EDITORIAL / OPINION

China moved first on national humanoid standards. On September 28, 2026, the State Administration for Market Regulation approved six GB/T documents that define technical requirements for general performance, environmental perception, decision-making and planning, motion control, task-oriented manipulation, and simulation testing platforms. They become effective April 1, 2027. The Beijing Humanoid Robot Innovation Center drafted the motion-control section while Unitree Robotics led environmental perception.

These documents are recommended national standards, not mandatory GB rules. In practice they function as de facto requirements for procurement, testing labs, and certification inside China. The series fills a gap that international bodies have left open. ISO/TC 299's safety standard for mobile robots with active stability remains a committee draft.

Lead organizations and drafting scale

Each of the six standards drew between 31 and 54 drafting organizations. Unitree Robotics headed the perception part, the Beijing Humanoid Robot Innovation Center handled motion control, UBTech Robotics took manipulation, and Pujiang National Laboratory covered decision-making plus the simulation platform. The Beijing Research Institute of Automation for Machinery Industry led the general-requirements section. This distribution mirrors the companies most active in China's humanoid deployments.

RobotToday reviewed the national standards database and counted 49 additional humanoid-specific national standard projects still in progress alongside 174 total active robotics and embodied-AI projects. One hundred five of those projects launched in 2026 alone. The pipeline includes matched test-method series, safety requirements split by use case, dataset specifications, component standards for joints and sensors, and operating-system and fleet-management documents.

Technical scope and concrete benchmarks

The simulation platform standard aims to eliminate inconsistent virtual environments that previously produced unreliable test results. Motion control and perception sections establish common terminology and performance thresholds that replace the current patchwork of company-specific metrics.

These choices matter for serviceability and cost. Uniform perception and control interfaces reduce integration time when swapping sensors or actuators across suppliers. Standardized simulation environments let developers compare results without rebuilding test rigs for each vendor. Over time that compresses commissioning labor and spare-parts inventory for operators running mixed fleets.

Timeline and next releases

The first six projects moved from launch in 2025 to publication in September 2026. Parts 6 and 7 of the technical-requirements series, covering localization and human-robot interaction, sit in technical review and are the likeliest candidates for release in the first half of 2027. Test-method documents aligned with each requirements part entered public comment in March 2026. Safety, dataset, and component series follow similar 12-to-18-month cycles, clustering around March-April and August-September 2027 publication windows.

Who writes the rules and foreign participation

UBTech appears on 32 humanoid drafting lists, Xiaomi Robotics on 26, and Unitree on 23. The Beijing Humanoid Robot Innovation Center, PaXini, Linkerbot, and telecom firms such as ZTE and China Mobile recur frequently. Foreign participation remains minimal. TÜV Rheinland and TÜV SÜD appear on a handful of projects; NVIDIA's Shanghai entity contributed to an ethics guidance document. No Western humanoid OEM sits on any humanoid drafting committee.

This pattern repeats a familiar dynamic. Chinese firms gain early influence over the metrics that define capability and safety. Buyers and regulators outside China may adopt the resulting test methods as reference points while ISO work continues. Companies shipping into the Chinese market will face GB/T 48317 benchmarks regardless of origin.

Geopolitical and market-access implications

China accounts for the world's largest robot market. Any humanoid or component supplier targeting volume sales inside the country will encounter these standards in tender specifications and certification schemes. The April 1, 2027 effective date overlaps with the EU Machinery Regulation's January 2027 application date, forcing dual-compliance work for firms active in both regions.

The standards also intersect with China's 10,000-unit national deployment plan and production scaling at firms such as AGIBOT and EngineAI. Uniform data formats and quality metrics across TC591, TC28, and TC609 dataset projects could enable pooled training corpora that fragmented Western efforts currently lack.

Counter-arguments and limitations

Recommended standards are voluntary on paper. Enforcement depends on how aggressively procurement offices and testing laboratories cite them. Early drafts may contain gaps that later revisions must close, especially around edge-case safety and long-term reliability metrics. Western firms retain the option to influence ISO processes and to differentiate on features that exceed the baseline.

Still, first-mover reference standards carry weight. Labs, insurers, and regulators often default to the most complete published framework when no international equivalent exists. The interlocking series structure—requirements, test methods, datasets, components—reduces the risk of contradictory rules that plague piecemeal approaches.

Supply-chain and labor effects

Standardized interfaces lower switching costs for integrators and end users. A factory running Tiangong units alongside Unitree or UBTech machines can share spare actuators and perception modules once common mechanical and electrical specifications stabilize. Training programs for technicians can cover a narrower set of diagnostic procedures rather than vendor-specific quirks. Over a multi-year deployment horizon those savi