South Korea has positioned itself to shape how the world defines and builds physical AI systems, while Chinese manufacturers are racing to deploy humanoid robots in factories within months. The competition reveals a critical moment: whoever writes the foundational standards for embodied AI now will influence every robot, autonomous vehicle, and smart factory built in the coming decade.

Why Does It Matter Who Writes the Physical AI Rulebook?

Physical AI, also called embodied AI, refers to artificial intelligence systems integrated into robots, autonomous vehicles, and industrial machines that must perceive and act in the physical world in real time. Unlike generative AI systems like chatbots that process information passively, physical AI systems must operate continuous perception-action loops running at 10 to 100 cycles per second. A robot catching a falling component on an assembly line must react within milliseconds; a delay that would go unnoticed in a chatbot would cause a physical failure.

The Electronics and Telecommunications Research Institute (ETRI), South Korea's government-funded research agency, has secured editorial control over the foundational vocabulary for the world's first United Nations-level standardization body dedicated to physical AI. The body, called the ITU-T Focus Group on Embodied AI for Multimedia Technologies (FG-EAI), was established in February 2026 by the International Telecommunication Union (ITU). This means 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 and autonomous systems has fully formed.

ETRI researcher Dr. Shin-Gak Kang chairs the working group responsible for foundational architecture and terminology, making him the designated editor of the Embodied AI Glossary. This glossary will define shared vocabulary that every other physical AI standard must cite. At the same time, ETRI division head Cha Hong-gi chairs the working group covering vertical industry applications across manufacturing, mobility, logistics, household automation, healthcare, and smart grids. Together, these positions give Korean researchers control over both the conceptual foundations and the sectoral implementation frameworks that other nations' engineers will need to follow.

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

Lee Kang-chan, Standards Research Division Head at ETRI

What Are Chinese Automakers Actually Building Right Now?

While Korea focuses on standards, Chinese automakers are moving faster on commercialization. Dongfeng Motor, one of China's largest automotive manufacturers, plans to begin small-batch trial production of its humanoid robot at the end of 2026, with the goal of achieving human-level working capability by the end of 2027. The robot will enter Dongfeng factories in October 2026, initially handling sorting and quality inspection tasks in manufacturing environments.

Zhang Zhenlin, chief engineer of intelligent technology at Dongfeng, explained that the company's in-house developed robot dog will enter commercial use sooner than the humanoid robot, first appearing in customer guidance scenarios at dealerships. Dongfeng's humanoid robot, called Xiaodong, is being continuously validated in material sorting scenarios and is evolving toward fully autonomous execution of flexible operations beyond single repetitive actions. The quadruped robot, Yuanzai, supports autonomous navigation, multimodal perception, and intelligent interaction, suited to campus patrol and smart service scenarios.

Dongfeng is not alone in this race. Xpeng, a Chinese electric vehicle maker, put a humanoid robot production line into operation on September 8, 2026, and reiterated that its Iron robot will begin mass production at the end of 2026, with plans to launch and deliver it in China and overseas in 2027. Aimoga Robotics, incubated by Chery, has already delivered more than 3,000 robots globally, including more than 2,000 overseas, and is preparing for an initial public offering to fund further expansion. Li Auto, Seres, and Leapmotor have also disclosed plans to enter the embodied intelligence field.

How Are Automakers Leveraging Their Existing Strengths in Robotics?

• Supply Chain Expertise: Chinese automakers possess decades of experience managing complex global supply chains for vehicle components, which they can now apply to sourcing and scaling robot hardware and sensors.
• Manufacturing Infrastructure: Existing automotive factories, production lines, and quality control systems can be repurposed or adapted to manufacture robots at scale, reducing time-to-market and capital investment.
• AI and Software Development: Automakers have invested heavily in automotive large language models and autonomous driving AI, which can be reused across robotics applications to accelerate commercialization and mass production.

Zhang Zhenlin noted that automakers' development of embodied intelligence products allows technology to be reused across multiple areas, thereby speeding up commercialization and mass production. He added that in the future, embodied intelligence products will not be limited to humanoid robots or robots in general; vehicles themselves will also be a form of embodied intelligence product.

What Technical Gaps Must Be Closed for Physical AI to Scale?

The practical engineering challenges facing the physical AI industry are substantial. Currently, physical AI systems from different manufacturers 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 gaps must be closed before physical AI systems can interoperate at industrial scale, including sensor protocols that 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, and actuator interfaces that govern how motor commands are transmitted.

This is precisely why the ITU-T Focus Group on Embodied AI exists. The governance window is open 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 structural positions held by Korean researchers mean they will shape how these interoperability challenges are defined and solved. 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.

The race between Korean standardization leadership and Chinese manufacturing speed reflects a broader competition in the physical AI space. While China accelerates deployment timelines and production capacity, South Korea is securing the conceptual and regulatory foundations that will govern how those systems operate globally. The outcome of this dual competition will shape not just which companies profit from physical AI, but how the technology is defined, certified, and integrated into industrial ecosystems worldwide.