Humanoid race may be won by the hand, not on foot

Earlier this month, Seoul’s COEX hosted an event whose name said it all: the China Humanoid Conference, “First Journey to Korea.” Unitree, Fourier, Leju and Huawei presented humanoids already working in Chinese factories and retail stores.

It felt less like an introduction than a declaration. The Korean engineers in the audience knew the scoreboard: China ships the vast majority of the world’s humanoids, with over 140 companies in the race. On volume and speed, the contest was over.

Not everyone on stage was triumphant, though. Yan Weixin, a senior researcher at Shanghai Jiao Tong University’s AI institute, acknowledged from the stage that his team had defined 33 fundamental hand motions and succeeded with 32. The exception: chopsticks, a task that demands “very precise force control and tactile perception.” Even China’s humanoids, Yan conceded, cannot yet perform the kind of delicate manipulation that most 5-year-olds master at the dinner table.

The distinction matters because it maps onto two very different competitions. Walking is an engineering problem China solved with speed and scale. Dexterous manipulation is a data problem no country has solved yet, and South Korean companies believe they hold some of the missing pieces.

China accounts for roughly 87 percent of global humanoid shipments, backed by over $26 billion in government funds, according to the Wall Street Journal. But Cho Eun-kyo, a research fellow at the Korea Research Institute who studies China’s advanced manufacturing, told The Korea Herald that “most factory deployments today involve logistics and patrol, tasks that rely on locomotion, not hands.”

A robot that can walk a factory floor but cannot grip a screw, insert a cable or fold fabric is, in commercial terms, an expensive courier. “The tasks that generate industrial value, that still account for 30 to 40 percent of an automotive production line, all live in the hand,” she said.

The technical reason for the gap is intuitive once you hear it. Choi Hyouk-ryeol, a mechanical engineering professor at Sungkyunkwan University and CEO of sensor firm Aidin Robotics, put it this way: close your eyes and pick up a paper cup. Your hand instantly registers whether it is empty, half-full or hot, and adjusts accordingly. Robots cannot do this. Their cameras see the cup but have no way of knowing how much force to apply.

Walking, Choi explained, is a “closed problem.” A flat factory floor is predictable, and mass data collection solves it. Manipulation is “open”: every object and surface demands a different grip, and the tactile data needed to teach that sensitivity barely exists. This is why China’s formula of rapid iteration at scale has not cracked the hand the way it cracked the leg, he said. It is also why Tesla reportedly keeps postponing factory deployment of its Optimus humanoid.

That unsolved problem is where Korean companies are concentrating.

Samsung Electronics reportedly established a dedicated “Hand Lab” within its Future Robotics Division in early March, adopting a tendon-driven mechanism that mimics human hand anatomy.

Robotis, a mid-sized Korean robotics firm, launched a five-finger hand this year at around 8 million won ($5,340), roughly 70 percent below competitors, by internalizing over 90 percent of components, including tactile sensors. Orders are backlogged for four months, with Google and Apple among early buyers, according to industry sources.

Beneath both efforts sits the sensing layer. Aidin Robotics, professor Choi’s own company, supplies force and torque sensors to 400 firms across 15 countries, including some in China. The moat, he argued, is not the hardware alone but the integrated “sensor, force control and AI solution” refined over years of deployment.

What connects these cases is a shared bet: the hand race will not be won on volume, but “on the fusion of mechanical precision, tactile sensing and data that only comes from real manufacturing environments,” Choi said. “Korea’s semiconductor fabs and automotive lines generate force and contact data that scarcely exists elsewhere.”

Professor Choi was also candid about the limits. “What Korea has is only a time advantage,” he said. Chinese firms were merely copying Aidin’s sensor designs two years ago, but “comparable products are now emerging.”

WeRobotics, a startup founded by two former Samsung engineers, knows this pressure firsthand. Its cable-driven hand system drew repeated visits from Nvidia, Amazon and Meta engineers at CES 2026, who assessed its dexterity as “world-class.” But CEO Kim Yong-jae said he is aware that the technology must reach production scale before Chinese competitors close the gap.

Cho at KIET said Korea should not try to match China’s full-stack humanoid ambitions. The realistic play is to become indispensable in a critical module.

“If Korea secures that position in the hand,” she said, “it doesn’t matter who assembles the rest of the robot.”

mjh@heraldcorp.com