Robot unified coding expands to all categories as embodied AI heats up in academia

Highlights

• MIIT plans "one machine, one code" for all robot categories; comment period runs to Dec 5
• Humanoid robot coding platform has covered more than 100 companies and over 28,000 units
• CoRL 2026 accepts 687 papers; Chinese institutions' share of authorship rises to 37%

Regulation: robot "ID cards" expand from humanoids to all categories

The First Equipment Industry Department of MIIT on Nov 6 opened public comment on the Interim Measures for the Management of Robot Product Identification Codes (Draft for Comment), with the comment period running to Dec 5. The core of the document is to establish a unified robot product "ID card" system, assigning every robot leaving the factory a unique product identity code to support industry safety governance and full-lifecycle management. The identification code consists of four parts: manufacturer code, product description code, product serial number code and other codes. Coding and code assignment must follow the national standard GB/T 48541-2026, Robot Product Identification Codes; the standard's development plan was issued in January 2026, it is administered by the National Robot Standardization Technical Committee, and it is currently in the comment stage. (Source: IT Home)

The measure does not start from zero. In May 2026, MIIT's Technical Committee for Humanoid Robots and Embodied Intelligence Standardization led the release of China's first humanoid robot full-lifecycle management service platform. The accompanying Specification for Humanoid Robot Full-Lifecycle Management uses a 29-digit, four-segment code — national code, enterprise code, product code and serial code — covering traceability across the entire chain from factory release and distribution to use and end-of-life recycling. As of May, the platform had covered more than 100 companies and over 28,000 humanoid robots had been assigned codes. (Source: IT Home)

The key change in the draft is the scope of application: from humanoid robots to all robot categories, with industrial, service and special-purpose robots all brought into the same coding framework. GB/T is a recommended national standard that companies may adopt voluntarily, but once it is cited in administrative regulations it acquires de facto binding force within their scope. For robot makers, product coding systems and traceability capabilities will gradually become a standard step before factory release, not an option. (Source: IT Home)

Academia: embodied AI is redrawing the research map

The heat in robotics academia is backed by concrete numbers. CoRL 2026 will be held Nov 9–12 in Austin, Texas, and has accepted 687 papers this year, a sharp increase from about 200 the previous year. Statistics show that from 2021 to 2025 the scale of CoRL papers grew at a modest linear pace (153 to 263 papers, an average annual increase of 14.5%), before jumping directly to 687 in 2026. Counting by whether a paper has authors affiliated with institutions in a given country, the share of US institutional authorship fell from 76% in 2021 to 51% in 2026, while China's rose from 5.9% to 37%. (Source: Leiphone)

Top conferences in the multimedia field are also adjusting their boundaries. ACM MM this year changed its call-for-papers topics from five major themes and 15 sub-directions to 16 peer-level research topics, adding two standalone themes: Embodied and Immersive Multimedia, and Multimedia Reasoning. The official definition of the former specifically mentions agents capable of perceiving, acting and interacting in the real world, and the integration of multiple types of information such as vision, language, audio and touch; but the conference stresses that this is multimedia and multimodal research within embodied intelligence, and does not involve robot hardware, control or motion planning. (Source: Leiphone)

Technical judgment: the bottleneck in physical intelligence lies between layers

Zhang Weinan, a professor at Shanghai Jiao Tong University's School of Computer Science and head of the Physical Intelligence Center at Shanghai AI Laboratory, said in a tutorial at CNCC 2026 that the core challenge of physical intelligence is enabling AI to form a complete loop of perception, prediction, action and continuous learning in a real world full of uncertainty where the consequences of actions are hard to undo. He laid out a seven-layer technology stack — from hardware and systems, data and simulation, to perception, world models, manipulation and motion control, and then to physical agents and industry applications — and stressed that each layer already has achievements; the truly difficult part often lies in the connections between layers, and in whether perception, decision-making, execution and feedback can be linked into a reliable whole. (Source: Leiphone)

Putting the three threads together: regulators are building a unified identity and traceability foundation for robots; academia is pushing embodied intelligence from a marginal topic into a standalone direction; and on the technology side, the difficulty of deployment lies not in single-point model capability but in system integration and continuous learning in the real world. All three point to the same thing — the robot industry is moving from "can it be built" to "can it be managed, evaluated and traced." The validation points for this judgment are: the coverage pace of the formal version of the identification code measures, the level of industry participation on site at CoRL 2026, and whether reusable engineering solutions emerge at the integration layers of the physical intelligence technology stack. (Source: Leiphone, IT Home)