China's AI showcase reveals impressive scale, visible limits

BEIJING — At a robot innovation center in the Chinese capital last week, a reporter tried speaking to the humanoid robot Tiangong 3.0 in Mandarin.

The robot — developed by the Beijing Humanoid Robot Innovation Center and billed as capable of understanding both Mandarin and English commands — had just finished a minute of dancing to "Moves Like Jagger" (2010), arms swinging, weight shifting with something close to rhythm.

"Let's shake hands," the reporter said. Tiangong 3.0 did not respond.

The request had to be repeated before it registered, and by the time the robot extended its arm, the moment had already made its point: the gait, when Tiangong moved between stations, had the same quality — deliberate, slightly mechanical, the walk of a machine that has learned to walk but has not yet forgotten that it is learning.

On June 29, the Korea JoongAng Daily visited three technology sites in Beijing as part of a Korean press delegation organized by China's Ministry of Foreign Affairs: the Beijing Autonomous Driving Pilot Zone in the Yizhuang economic development zone, the Humanoid Robot Innovation Center and LLVision, a smart glasses company.

China's AI sector is advancing at genuine scale. The gaps, on close inspection, are also real.

Self-driving: A bus with no one at the wheel

The autonomous bus pulled away from the curb smoothly — no jerking, no hesitation — and merged into traffic at a pace that felt, within a few seconds, unremarkable.

There was no driver's seat. No one stood at the front. The vehicle navigated lane changes, traffic signals and a roundabout with an eerie, calculated confidence.

The vehicles in the Yizhuang pilot zone operate on a principle that separates Chinese development from the American approach.

Waymo and Tesla concentrate computing intelligence in the vehicle itself, training systems to read and react to the environment around them.

Beijing's model pushes intelligence onto the road.

Sensors, signals, holographic devices and smart poles embedded throughout the pilot zone feed real-time data to the vehicles traveling through them, compensating for limits that any individual vehicle's sensors will always have: strong sunlight that obscures traffic lights, obstacles that create blind spots, the inability of one car to anticipate the whole flow of traffic around it.

Company representatives called it "vehicle-road integration," or V2X, and described it not as a workaround but a design choice — one that works well inside a zone built to support it and raises questions about scalability anywhere the infrastructure has not been laid.

For a reporter familiar with Korea's own autonomous driving testbeds, such as Sangam-dong in Mapo District, western Seoul, the contrast in philosophical scale was stark.

While Sangam-dong pilots autonomous buses and RoboShuttles, those vehicles must constantly battle Seoul's famously cramped, legacy urban architecture, relying heavily on their own onboard cameras and light detection and ranging.

Yizhuang, by contrast, felt like an entire city grid systematically reengineered from the asphalt up specifically to accommodate the machine.

Level 4 autonomy, high driving automation where a vehicle can handle all driving tasks entirely without human intervention, which the vehicles in Yizhuang represent, "has not yet reached full commercial deployment," one representative said. What journalists rode was advanced technology in a controlled setting.

On a competitive scale, however, the official said Pony.ai, WeRide and Baidu's Apollo Go have deployed close to 5,000 Level 4 vehicles combined across China. Waymo has more than 3,000 vehicles in the United States. No other country is in the conversation.

Humanoid robots: The gap between the reel and the room

Without warning, Tiangong 3.0 dropped to its hands and knees and began crawling.

A low obstacle course had been set up across a gravel surface, and the robot moved through it methodically, its joints absorbing the uneven ground — emerging on the other side with visible wear marks on its knees that no one in the room commented on.

It went on to move boxes between shelves at different heights, scan bar codes, carry a drink across a room, unlock a door, retrieve a package from outside a simulated home and return to its charging station on command. Its handshake, a representative noted, exerted actual force.

The robot was developed through joint investment by companies including Ubtech and Xiaomi Robotics, and completed a half-marathon in 2025 — a milestone the center's materials featured prominently, though the race involved multiple robots and several substitutions along the way.

Center officials described near-term targets as manufacturing floor deployment, with elder care being the long-term ambition.

The comparison to Tesla's Optimus was implicit throughout. No side-by-side technical benchmarks were offered.

Korea has a stake in how this competition plays out.

Hyundai Motor's ownership of Boston Dynamics puts a Korean company at the center of the global humanoid race. The Beijing center's representatives, pitching manufacturing deployment as their near-term goal, are essentially targeting the same factory floors where Korean industrial robots and Korean-affiliated automakers currently operate.

AI glasses: 12 years on one problem

Ma Qingyang, LLVision's marketing director, was speaking in Mandarin.

To the journalists in the room wearing the company's Leion Hey2 glasses, green Korean text was appearing at the bottom of their field of view in real time, keeping pace with his words close enough to follow the argument — though the translation at times felt slightly rigid.

LLVision has been focused on smart eyewear since 2012, incorporated in 2014, and spent the intervening years developing what Ma described as a new product category: subtitle glasses.

The Hey2, shown at CES 2026, projects real-time translated subtitles as the wearer listens to speech in a foreign language — supporting, the company says, more than 100 languages.

The company has collected recognitions including a Unesco global innovation award and selection by Harvard Business Review as one of the 10 technology trends of 2024, alongside OpenAI's ChatGPT.

Ma cited Meta's more than $150 billion investment in augmented reality as evidence that smart glasses are headed toward replacing the smartphone. LLVision's claim is that language translation is the application that gets there first.

For Korean companies, that claim carries a direct implication.

The simultaneous interpretation market — dominated by specialized hardware from companies including Seoul-based manufacturers and localized translation apps — is one of the sectors LLVision is positioning to enter.

A product that puts real-time translation on a pair of sleek glasses at consumer price points does not just threaten American tech platforms. It lands closer to home than it might first appear.

BY SEO JI-EUN   [[email protected]]