Etri.re.kr

South Korea's national research institute has secured editorial control over the foundational vocabulary of the world's first United Nations-level standardization body for Physical AI — meaning Korean researchers will write the first drafts of definitions that every future international Physical AI standard must build upon, before the global market for robots, autonomous vehicles, and industrial AI has fully formed. ITU's FG-EAI official overview

The body in question is the ITU-T Focus Group on Embodied AI for Multimedia Technologies (FG-EAI), FG-EAI established February 2026 by ITU-T Study Group 21 on February 19, 2026 — the first ITU-T focus group dedicated exclusively to Physical AI governance. The Electronics and Telecommunications Research Institute (ETRI), South Korea's government-funded technology research agency, proposed the body in late 2025 in partnership with Hankuk University of Foreign Studies, and then moved quickly to occupy its most structurally significant positions once the group convened.

What Physical AI Is, and Why Governing It Now Matters

Physical AI — sometimes called embodied AI — is artificial intelligence integrated into systems that perceive and act in the physical world: robots, autonomous vehicles, humanoid machines, and smart factory systems. The term encompasses any AI that must close the loop between sensing and actuation in real time, not process data passively in a cloud. Physical AI differs from generative AI

The technical distinction from generative AI is architectural, not cosmetic. A generative AI system — a large language model or image generator — performs stateless inference: text goes in, output comes out, and there are no physical consequences if the process takes two seconds. Physical AI systems must operate a continuous perception-action loop running at 10 to 100 cycles per second. Embodied AI's closed-loop control requirement A robot catching a falling component on an assembly line must react within milliseconds; a latency that would go unnoticed in a chatbot would cause a physical-world failure. This constraint — real-time closed-loop control under strict latency, power, and safety limits — is what makes Physical AI technically distinctive, and what makes standardizing it significantly harder than standardizing cloud AI software.

The governance window is open precisely because Physical AI remains in early commercialization. Standards written now will determine what hardware can legally interoperate in certified environments, which certification regimes companies must satisfy before deployment, and which network quality-of-service guarantees telecommunications infrastructure must provide to AI-driven physical systems. A technology ecosystem built on incompatible proprietary protocols becomes commercially fragmented; a technology ecosystem built on shared standards becomes a single addressable market. The standards race shapes future market access for the country whose researchers write the vocabulary of those standards.

Korea Secures Commanding Heights in Working Group Structure

When FG-EAI's second full plenary convened in Geneva on July 11, 2026 — alongside the ITU-T Study Group 21 session that drew 509 participants from 35 countries — the working-group leadership structure was formally constituted, and ETRI emerged with a position set that spans both conceptual foundations and applied deployment. FG-EAI management team composition

FG-EAI's chair is Yuntao Wang from CAICT, the China Academy of Information and Communications Technology, a state research entity under China's Ministry of Industry and Information Technology. Two vice-chairs balance that appointment: ETRI researcher Dr. Shin-Gak Kang and Avinash Agarwal, Deputy Director General at India's Ministry of Communications Department of Telecommunications.

Dr. Kang also chairs WG1, the working group responsible for foundational architecture and terminology — the group that will produce the shared vocabulary every other working group's output must be consistent with. At the same time, ETRI division head Cha Hong-gi chairs WG6, which covers vertical industry applications across manufacturing, mobility and logistics, household automation, healthcare, and smart grid. Researcher Hyun Wook chairs TG6.1, the task group specifically focused on manufacturing environments. Professor Jeong Seong-ho of Hankuk University of Foreign Studies chairs TG1.2, which is developing the reference architecture — the full-stack engineering blueprint — for Physical AI systems. FG-EAI working group leadership

Holding both WG1 and WG6 simultaneously is the structural fact that makes this more than a committee seat. WG1 produces the shared definitions every standard must reference; WG6 determines how Physical AI should be specified for deployment across every major industrial sector. Whoever chairs WG1 sets the terms of the conversation; whoever chairs WG6 determines the use cases that define which features the standards must accommodate. Together, they cover the arc from first principles to sectoral implementation.

Why Editorship of a Glossary Is a Power Position

In ITU-T standards work, Focus Group outputs — Technical Specifications and Technical Reports — are written by designated editors who draft the initial text for review and revision by the working group. ITU-T Focus Group mechanism explained First drafts in standards negotiations have enormous staying power: they establish the conceptual framework and vocabulary from which all subsequent negotiation departs, and they set defaults that remain in place unless a participant coalition actively overrides them. An editor who defines a term does not get to dictate the final standard, but does get to establish what everyone else must argue against.

ETRI researcher Dr. Kang is the designated editor of the Embodied AI Glossary (formally: TS-Term-glossary), the FG-EAI Work Programme document deliverable that will define the shared vocabulary every other Physical AI standard must cite. Whoever writes the first draft of "Physical AI," "embodied agent," "closed-loop control," and "multimodal sensor fusion" in this context shapes what those terms mean in every certification framework, every procurement specification, and every regulatory scheme that references the resulting standard. Professor Jeong's team separately edits the Reference Architecture document that translates those definitions into an engineering framework. The combination of lexical authority (glossary) and technical architecture authority (reference architecture) means Korean researchers hold both ends of the conceptual apparatus that other nations' engineers will need to implement.

"Physical AI is recognized as a key technology that will transform the paradigm of manufacturing, mobility, healthcare, and service industries following generative AI," said ETRI Standards Research Division head Lee Kang-chan. "The country that leads in international standards will lead the future industrial ecosystem."

What Standardization Must Actually Solve

The practical engineering challenges FG-EAI is chartered to address are real and documented. Physical AI systems from different manufacturers currently operate on incompatible protocols: a depth camera from one vendor cannot natively communicate its output to another vendor's planning software in a standard format; an actuator's joint-position reporting uses proprietary encoding that no cross-vendor standard governs. Five categories of gap must be closed before Physical AI systems can interoperate at industrial scale.

Sensor protocols define how perception data from cameras, depth sensors, tactile arrays, and LiDAR gets formatted and transmitted so AI planning layers from any vendor can ingest it. Actuator interfaces govern how motor commands are sent and how joint-position data is reported back. Network latency guarantees specify what telecommunications infrastructure must provide — cloud AI tolerates round-trip latency above 100 milliseconds; Physical AI control loops typically require under five milliseconds end-to-end, and no existing network QoS standard addresses that constraint. Safety certification frameworks establish what testing a Physical AI system must pass before deployment in a factory, hospital, or public environment — frameworks that currently differ by country, sector, and regulatory body. Cross-vendor interoperability specifications determine whether a robot from manufacturer A, using sensors from manufacturer B, running inference on edge compute from manufacturer C, can legally operate as a certified system in a shared environment.

FG-EAI has organized working groups around these gaps. WG1 handles foundational architecture and terminology. WG2 covers multimodal data frameworks. WG3 addresses AI models. WG4 handles systems integration and interfaces — the actuator and sensor protocol layer. WG5 covers connectivity and interoperability — the networking layer. WG6 handles vertical industry applications. WG7 addresses human alignment and interaction, with a task group specifically covering accessibility. WG8 focuses on evaluation and benchmarking. FG-EAI eight working groups

The group has already issued liaison statements to more than 20 peer standardization bodies, including ISO/IEC JTC1/SC42 (AI), IEC TC 65 (industrial process measurement), ISO TC 299 (robotics), IEEE RAS (Robotics and Automation Society), and ISO TC 204 (intelligent transport systems) — covering every major domain where Physical AI will operate. FG-EAI liaison statements to peers

How Korea Got Here: A Broader Standards Strategy

The FG-EAI positions are the most visible element of a deliberate South Korean government strategy to shift from technology fast-follower to standards first-mover across AI domains. In November 2025, ETRI established the first ISO/IEC international standard for AI system testing to be led by South Korea — the Overview of AI System Testing (ISO/IEC TS 42119-2) — after more than five years of work. ETRI led first AI testing standard In April 2026, a separate group established the AI Advisory Group within ISO/IEC JTC 1/SC 6, with a Korean expert as co-convener alongside China. Korea co-leads ISO AI advisory group In June 2026, Korea's standards agency signed a memorandum of understanding with Singapore, the United Kingdom, Australia, and Canada on AI pre-standardization cooperation. Korea's AI pre-standardization MOU

The governmental investment backing this effort is substantial. Korea's Sixth National Standards Master Plan, a five-year blueprint covering 2026 through 2030 developed jointly by the Ministry of Trade, Industry and Resources and seventeen co-ministries, commits approximately KRW 1.49 trillion (approximately $1.08 billion USD) to standardization activities. Korea's Sixth Standards Master Plan Physical AI is among the 18 priority sectors the plan identifies for international standardization, alongside quantum technology, semiconductors, and next-generation communications.

The Geopolitical Stakes in Standards Competition

The race to govern Physical AI at the international standards level is not a technocratic administrative process. Across the broader technology standardization arena, the US Department of Commerce has estimated that 93% of global trade affected by standards — specifications that determine which products may legally be sold, certified, and deployed in which markets.

China has invested heavily in technical standards leadership across ITU, ISO/IEC, and 3GPP over the past decade, using state subsidies to fund standards participation at a scale private-sector companies in other countries cannot match unilaterally. A 2021 Carnegie Endowment study found that US industry views China's SDO push raised competitive concerns — most commenters did not believe China was violating SDO rules, but did believe that US competitiveness was threatened by the way China was mobilizing state resources around standards-setting. A January 2025 Asia Society policy brief recommended that the Trump administration establish a "Standards Alliance" specifically with Japan, South Korea, Germany, and the United Kingdom to coordinate standards positions and counter China's state-backed mobilization. Asia Society Standards Alliance recommendation

Within FG-EAI specifically, the leadership structure reflects that geopolitical reality. China's CAICT holds the chairmanship. Korea holds more working-group-level positions than any other country, plus editorial authority over the two most foundational deliverables. India's government holds the second vice-chairmanship. The structure requires coalition-building and consensus — that is the ITU's design — but editorial authority over first drafts is not determined by consensus votes. It is determined by who is in the room writing.

What Comes Next

FG-EAI working group sessions are proceeding on an intense inter-session schedule. WG1/TG1.1 — the terminology task group where the glossary is being drafted — will meet virtually on September 21. WG6/TG6.1, covering manufacturing applications where ETRI's Hyun Wook leads, will meet September 28 through 29. The group's third full plenary is scheduled for October 13 through 16 in Hangzhou, China. FG-EAI upcoming meeting schedule

The focus group's outputs — Technical Specifications and Technical Reports — are not yet binding ITU Recommendations. The ITU-T mechanism is designed to accelerate work on fast-moving topics, with the expectation that Focus Group deliverables will be proposed into the parent Study Group for progression into formal Recommendations when the evidence base is sufficient. ITU Focus Group outputs pathway The Physical AI governance architecture being built now will, if the process runs its course, become the binding framework against which future robot systems, autonomous vehicle platforms, and industrial AI deployments are certified worldwide.

Korea's researchers currently hold the pen on the first draft of what that framework calls things, and what it requires them to do.

Frequently Asked Questions

What exactly is the ITU-T Focus Group on Embodied AI, and does it make rules companies must follow?

FG-EAI is a subsidiary body of ITU-T Study Group 21, the UN telecommunications agency's division covering multimedia and content delivery technologies. Focus Groups are an ITU-T mechanism for accelerating standards work in fast-moving technology domains — they can participate freely (membership is open, unlike some Study Group processes), and they produce Technical Specifications and Technical Reports. Those outputs are not yet binding international standards, but they serve as the foundation texts when the parent Study Group later considers elevating them to formal ITU-T Recommendations, which member states may then incorporate into national law. Think of FG-EAI's outputs as the first authoritative drafts of what will likely become binding standards for Physical AI certification, interoperability, and safety.

What does it mean in practical terms that ETRI will write the Physical AI glossary?

In standards development, the editor of a foundational glossary writes the first working draft of every key definition — what "embodied agent" means, what "closed-loop control" requires, what qualifies as a "Physical AI system" for certification purposes. First drafts in standards negotiations carry significant weight because they establish the conceptual baseline from which all subsequent discussion departs. Participants can propose revisions, and consensus governs the final text — but the editor sets the terms of the conversation. For companies building robots, autonomous vehicles, or industrial AI systems, these definitions will eventually determine what their products must demonstrate to achieve international certification.

Why doesn't the United States have leadership positions in FG-EAI?

That is a question the US standards community is actively discussing. A 2025 Asia Society policy brief recommended that the Trump administration establish a formal "Standards Alliance" with Japan, South Korea, Germany, and the United Kingdom to coordinate standards positions — precisely because state-backed actors like China and, in this case, Korea's government-funded ETRI are able to sustain the intensive, years-long standards participation that private-sector US companies find difficult to fund unilaterally. The US standard on this is that private industry leads; the gap is that other governments subsidize participation directly.

How does Physical AI standardization affect the companies that build or buy industrial robots?

Once Physical AI interoperability standards mature into binding ITU Recommendations, they will determine what hardware combinations can legally operate in certified industrial environments. A factory deploying a robot system will need to demonstrate that its components — sensors, compute, actuators — comply with certified interfaces. Systems built on proprietary protocols may face market access barriers in jurisdictions that adopt the ITU standards. For buyers, the practical effect is that certified Physical AI systems from different vendors will eventually be able to interoperate on shared factory floors, reducing vendor lock-in. For builders, the standards race determines whose technical architecture becomes the reference point.