China now has official standards for grading how capable a humanoid robot is and handling its training data

China’s industry ministry has published six humanoid robot standards, including a scale for grading a robot’s intelligence and rules for its training data.

19 Nov 20262 min readPolicyRoboticsChina
6: humanoid robot standards, including a scale for grading robot intelligence

China’s industry ministry has published six standards for humanoid robots and embodied AI, covering how the robots are trained, how their intelligence is graded, and how key parts such as hands and joints should perform and be tested.

The Ministry of Industry and Information Technology (MIIT) approved them in Announcement No. 30 of 2026, issued on 10 November and published on 16 November, as part of a batch of 274 industry standards. The six take effect on publication. They carry “SJ/T” numbers, the label China uses for recommended rather than mandatory electronics-sector standards.

One, SJ/T 12355, sets a framework, level definitions and assessment methods for grading a humanoid’s intelligence, for use in research, testing and when choosing a robot for a job. Another, SJ/T 12354, sets rules for training data across its whole life, from collection to destruction. The others cover how to build robot training grounds (including safety and ethics), systems that generate embodied-AI training data, dexterous hands, and motor-driven joints.

Why it matters

humanoid makers describe their robots’ abilities in very different ways. A shared grading scale and common test methods give buyers and officials in China a way to compare them. Data rules matter because today’s robots learn largely from recorded human demonstrations, often collected by remote control.

Related on AIooze: Embodied AI and Teleoperation in the Robotics glossary.


Related