LG Electronics has raised its robot-training data target to approximately 500,000 hours and said preparations for actuator mass production are progressing. Management outlined the plans on its 28 October third-quarter earnings call, alongside a fourth-quarter opening for the Seoul Data Factory. The cumulative data target combines real-world recordings with synthetic and augmented material.
The company’s August collaboration announcement with NVIDIA set a 100,000-hour objective for the 10,000 m² Yangjae facility. Its training environments reproduce household cleaning, manufacturing and logistics tasks, including processes based on LG’s Tennessee washing-machine plant. Several hundred robots are planned for the site, with collected data feeding development of LG’s robot foundation model.
LG uses NVIDIA Omniverse libraries, Isaac robotics tools and Cosmos world models to connect physical task data with simulated and generated examples. The facility includes areas for validating and refining datasets as robots repeat tasks across manufacturing and home environments.
LG also established an actuator pilot line in Changwon in the first half of 2026. Management said it is adding automation and quality controls as it prepares larger-scale production and pursues external component customers. Its initial commercial focus spans manufacturing and logistics, with expansion into home humanoids planned from around 2030.
The Data Factory’s immediate role is to improve LG’s own robot models and products. Management is considering paid pilots, dataset development and data services for external customers as longer-term business opportunities.
