"When you look at real work sites, no single humanoid can handle every task," Joo Si-hyun, vice president of the robotics intelligence development group at Hyundai Motor Group, told the audience of 2026 GGGF in Seoul Thursday.
The industry's fixation on human-shaped machines obscures the wider family of physical AI systems doing the actual work, a point that applies well beyond any single carmaker.
Physical AI refers to systems that act in the physical world rather than on a screen, steering robots, vehicles and machines through real space.
It is widely cast as the next frontier after generative AI, and a humanoid form is often treated as its natural endpoint.
Yet a humanoid cannot cover every task anindustrial site demands, Joo said. It can climb stairs and strike a range of poses, but its payload and speed are limited, while a single plant can span indoor corridors, outdoor yards and staircases.
Stairs suit legged robots of the kind exemplified by Boston Dynamics' Atlas and Spot, while flat factory floors without curbs favor wheeled autonomous mobile robots that glide between stations.
Behind a single physical AI machine sits a stack of technologies that never appears in a demonstration, spanning embedded intelligence, perception, autonomous driving and fleet management, each a discipline of its own.
The field is pushing toward end-to-end systems that fold those layers into a single model, he said, but technical bottlenecks remain, forcing a compromise between what is realistic now and where the technology may eventually go.
Cost is another constraint the sector cannot dodge.
Squeezing prices by fitting cheaper parts and mass-producing them, the route Tesla has taken, is not what buyers actually want, Joo argued, since no customer asks for inferior components.
Standardizing shared parts matters more, and the motors and actuators that give a robot its movement remain among the most expensive elements.
The near-term aim is machines anyone can operate, including the elderly, on platforms open enough for others to build services on top.
A follow-me function that trails a factory worker, or a golf caddie that shadows a player, hints at the range. The group is also developing electric-vehicle charging robots for high-speed cables that can weigh 5 to 10 kilograms, a burden for older or disabled drivers.
"I believe robotics can solve a wide range of social problems, but no one group can do it alone," Joo said, closing his address with a call for partners across components, services, insurance and communications.
His call for collaboration set the tone for the session that followed.
Lim Kyeong-seob, who heads manufacturing-AI transition cooperation at the Ministry of Trade, Industry and Resources' industrial AI policy bureau, used his talk on the M.AX Alliance to argue that the manufacturing base long credited with driving South Korea's growth has hit a ceiling.
"The competitiveness of manufacturing, which has driven our economy, has reached its limit," Lim said.
"We have to take a different path from the United States and China, and we believe Korea needs a private-sector ecosystem where manufacturers and AI firms work together — which is why the M.AX Alliance was created."
Later sessions turned to the token economy's vanguard and the rise of the AI factory, before the forum closes Friday with a private session at the National Assembly.
The three-day event, running under the theme "From Physical AI to Agentic AI, Beyond to AI Civilization," opened Wednesday at The Plaza in central Seoul, hosted by Aju News Corporation with the Global Economic and Financial Research Institute.
AJP Takeaways
• Hyundai Motor Group VP Joo Si-hyun told the 2026 GGGF that humanoids dominate the physical-AI conversation but can't cover most industrial tasks — wheeled AMRs and legged robots will hit factory floors first.
• A single visible robot rests on an unseen stack — perception, embedded intelligence, autonomous driving and fleet management — split between optimal control and learning-based control, with true end-to-end systems still bottlenecked.
• Cost hinges on standardizing shared parts rather than cheaper components, and no firm can build the full system alone, Joo said, citing Hyundai's charging robots and follow-me machines as near-term uses.
Copyright ⓒ Aju Press All rights reserved.