LG Electronics and NVIDIA are accelerating their robotics partnership by building one of the world’s largest robot training datasets, targeting 100,000 hours of robot training data by the end of 2026.
The initiative is centered around LG’s new Data Factory in Seoul, a dedicated robotics training facility where hundreds of LG CLOiD robots will continuously perform household, logistics, and manufacturing tasks. The collected real-world data will be combined with synthetically generated data created using NVIDIA Cosmos to train LG’s next-generation Robot Foundation Model (RFM).
The 100,000-hour target is equivalent to roughly 12 years of continuous robot operation, making it one of the largest physical AI datasets assembled by a consumer electronics company. The facility will include simulated homes, manufacturing lines modeled after LG’s Tennessee appliance factory, logistics environments, and robotic manipulation labs to expose robots to a wide variety of real-world scenarios.
The partnership combines LG’s decades of manufacturing and logistics data with NVIDIA’s robotics software stack, including Omniverse, Cosmos, and the Isaac robotics platform. Together, the companies aim to create a continuous “data flywheel,” where robots collect data, AI models improve from that data, and better-performing robots generate even more valuable training data.
LG plans to use the resulting foundation model across a broad portfolio of robots, ranging from industrial automation and logistics systems to future home robots and humanoid platforms. The company recently established a dedicated Robotics Business Center reporting directly to the CEO as it expands beyond appliances into physical AI.
The collaboration follows a strategic agreement signed by LG Group and NVIDIA just days earlier and reflects the industry’s growing focus on physical AI. As humanoid robots move closer to commercial deployment, access to large-scale, high-quality training data is increasingly viewed as one of the industry’s most valuable competitive advantages.
