South Korea’s AI and robotics push came to the fore again after LG Electronics hosted Nvidia senior director Madison Huang during a Tuesday trip to its robot data factory in southern Seoul.

The younger Huang’s visit to the Yangjae campus in Seocho District came after her father, Nvidia CEO Jensen Huang, met and signed an MOU with LG Group Chairman Koo Kwang-mo covering robotics, AI factories and mobility in Santa Clara, California. Madison Huang also handles product marketing for Nvidia’s physical AI platform, the company’s term for AI models that let machines sense and act in the physical world.

Nvidia commits to South Korea’s robotics push

According to local reports, Madison Huang was received by C-suite executives including LG Electronics CEO Lyu Jae-cheol, LG CNS CEO Hyun Shin-gyoon and LG Sciencepark’s chief. The visit began with a courtesy bouquet delivered to Huang by one of the company’s CLOiD robots and ran longer than the two hours originally put on the schedule.

Huang described her visit to the facility as “incredible,” doubling down on the positive sentiment she expressed when she left a handwritten “Amazing LG” on a CLOiD robot.

100,000 hours of training data by December

LG expects to have its robots running on Nvidia’s training platform before the end of 2026, with plans to have at least 100,000 hours of robot training data in its library by the end of the year.

The plan will combine physically tracking robots doing tasks with real on-site motion data. Nvidia’s Omniverse libraries, LG’s Cosmos world foundation models and the Isaac development platform will then be used to amplify and synthesize the data into far larger virtual datasets, which feeds LG’s robot foundation model.

That emphasis on data reflects a central bottleneck in robotics: unlike text, real-world robot motion data is slow and costly to collect, which is why developers increasingly use simulation to multiply limited real-world demonstrations into large synthetic training sets.

LG Electronics CEO Lyu Jae-cheol’s team described the combined pool as the robot’s brain, which lets it handle objects even though it does not have specific training for such tasks. That self-improving loop of collection, synthesis, training and improvement is what LG describes internally as a “data flywheel.”

Inside the Yangjae robot data factory

Madison Huang visited the 10,000-square-meter converted Yangjae R&D campus in Seocho District. The facility is expected to be running hundreds of CLOiD robots on a 24/7 schedule by the end of the year.

LG has simulated environments it expects to deploy its humanoid robots in across the facility’s four floors, including one below ground:

• CLOiD practices cleaning in the home zone.
• Robots move, stack and assemble washing machine parts in a setting resembling the assembly line at LG’s Tennessee plant.
• LG CNS added a logistics-automation area.
• LG Innotek operates a space for training robotic hands.

LG has ambitions to build a bipedal, next-generation humanoid on Nvidia’s robotics stack before the end of March 2027. The South Korean firm named other Nvidia projects in its Friday report on H1 2026.

The timetable places LG in an increasingly crowded humanoid field that includes Tesla’s Optimus program, Figure AI and Hyundai Motor Group’s Boston Dynamics, while at home Samsung Electronics has been building out its robotics presence through Rainbow Robotics.

“We will secure competitiveness in physical AI,” Lyu said, adding that LG aims to become “a total robotics solutions provider.”

Seoul’s wider physical AI bet

The partnership sits inside a national push. President Lee Jae Myung’s government has folded physical AI into a roughly $1 trillion program alongside semiconductors and AI data centers, and plans a special law this year to fast-track approvals and infrastructure.

Industry Minister Kim Jung-kwan said the state intends to buy more than 1,000 robots by 2030, including 700 humanoids, to seed an early market that Lee has said is still too small. Before then, the calendar already holds nearer-term markers: the special law planned this year, LG’s 100,000-hour data library by December and the humanoid target before March 2027.