LG and NVIDIA are expanding their robotics partnership. The new Data Factory at LG’s research campus in Seoul is expected to gather approximately 100,000 hours of training data for robots by the end of 2026. With this, LG aims to accelerate its Physical AI strategy and make the leap from industrial robots to household robots.
Following the joint memorandum of understanding signed on August 13, 2026, in Santa Clara, executives from LG and NVIDIA met in Seoul to finalize the next steps. However, the collaboration between the two companies is not entirely new: Back in June, NVIDIA and the LG Group had already announced a significantly broader partnership centered on a joint AI Factory, which encompasses not only robotics but also autonomous driving and AI infrastructure. The current agreement essentially constitutes the robotics chapter of that partnership, now with a concrete roadmap.
In short:
- LG and NVIDIA are deepening their robotics collaboration through the Data Factory at the LG Research Campus in Seoul.
- By the end of 2026, 100,000 hours of real and synthetic training data for robots are expected to be generated there.
- NVIDIA Omniverse, Cosmos, and Isaac (including GR00T) will be used for simulation and robot control.
- The data will be integrated into LG’s Robot Foundation Model and the CLOiD home robot.
- LG established its own Robotics Business Center in July 2026 and aims to expand its presence across the spectrum from industrial to household robots.
What is the LG Data Factory in Seoul?
The Data Factory is located on the LG research campus in Yangjae, Seoul, and spans four floors with a total area of approximately 10,000 square meters. There, LG’s CLOiD home robots are trained under conditions that are as realistic as possible: There is a simulated home environment for cleaning tasks, a manufacturing simulation modeled after LG’s washing machine plant in Tennessee, and an area for logistics automation provided by LG CNS. LG Innotek is also contributing its own training area for robotic grippers.
By the end of the year, several hundred robots are expected to be in operation there, collectively accumulating around 100,000 hours of real and synthetically generated training data. According to LG, this is equivalent to about twelve years of accumulated experience—an impressive figure, even if it remains unclear exactly how training hours translate into real-world experience. One thing is clear: a cycle is being created. Robots collect data from real-world operations; that data is processed and supplemented with synthetic scenarios; and both feed back into the training of the next generation of robots.
These NVIDIA technologies are behind it
For training and simulation, LG relies on several components from the NVIDIA robotics stack. The following table shows what each platform is used for:
| NVIDIA Technology | Purpose at LG |
|---|---|
| NVIDIA Omniverse | Simulation and digital twins of training environments |
| NVIDIA Cosmos | Generation of synthetic training data (world models) |
| NVIDIA Isaac (Sim, Lab, GR00T) | Development platform for robot control and humanoid base models |
These tools are designed to enable LG to train robots not only in the real-world Data Factory but also, in parallel, on a large scale in simulation—significantly faster and more cost-effectively than would be possible with physical test runs alone.
CLOiD and the Robot Foundation Model
At the heart of the Data Factory is LG’s own household robot, CLOiD, which the company had already unveiled at CES 2026. The data collected there feeds directly into the further development of LG’s Robot Foundation Model (RFM), a base model that, according to LG, is designed to improve the capabilities of humanoid robots. The idea behind it is similar to that of a large language model, except that in this case, the system learns movement sequences instead of words. According to reports, LG is aiming for the first quarter of 2027 to launch its first bipedal humanoid robot, though this date has not yet been officially confirmed.
From Industrial Robots to Household Robots: LG’s 2026 Robotics Initiative
LG has declared 2026 as the starting point for its robotics initiative. Building on decades of experience in manufacturing and logistics, the company now plans to specifically expand its portfolio of industrial and service robots to include household robots. Vacuuming and mowing robots have long been part of everyday life in many households, as recently demonstrated by tests of the Beatbot Sora 70 pool robot and the Yarbo lawn mowing robot. Unlike these specialized devices, CLOiD is intended to be a versatile assistant for various household tasks—at least that’s the plan.
To this end, LG established its own Robotics Business Center in July 2026, which reports directly to the CEO and coordinates the company-wide robotics strategy—from individual components such as actuators to complete hardware and software solutions.
Context: The Race for Physical AI
LG is not alone in its robotics ambitions. Samsung and the SK Group are also advancing their own Physical AI programs in South Korea, while NVIDIA, with Isaac and GR00T, is becoming the preferred infrastructure partner for humanoid and mobile robots across industries. LG Electronics CEO Lyu Jae-cheol summed up the strategy as follows: “Through the synergy we’re harnessing under the ‘One LG’ motto and through strategic partnerships with global partners, we will secure our competitiveness in the field of Physical AI.”
Conclusion
With the LG-NVIDIA collaboration and the Data Factory in Seoul, LG is creating the data infrastructure necessary for competitive Physical AI products. Whether this will actually result in a market-ready humanoid robot by 2027 remains to be seen. However, the roadmap—which includes 100,000 training hours by the end of the year and the establishment of its own Robotics Business Center—shows that LG is now pursuing robotics with the same level of commitment as its core business of home appliances and consumer electronics.
