LG, Nvidia Spar in Robot Training Data Arms Race
LG Electronics and Nvidia have entered into a high-stakes competition to generate 100,000 hours of robot training data at LG's DataFactory in Seoul. This ambitious goal surpasses the existing benchmark of 60,000 hours set by Ant Group Robbyant for its LingBot-VLA 2.0 model.
The DataFactory is a state-of-the-art facility that simulates real-world environments, allowing robots to perform structured tasks and collect sensor data. The facility's design enables LG to anchor on real manufacturing-environment data, which it then amplifies using Nvidia's Cosmos open world models to generate synthetic variations.
LG plans to deploy several hundred CLOiD units at the DataFactory by year-end, with the goal of reaching 100,000 hours of training data. This approach leverages LG's manufacturing legacy data and combines it with Nvidia's robotics software stack, which is expected to be a key differentiator for the company.
The joint effort between LG and Nvidia aims to accelerate the development of advanced robots and AI infrastructure. The companies have committed to delivering several key projects, including a next-generation bipedal humanoid robot built on Nvidia's Isaac GR00T foundation model, which is scheduled for public unveiling in 2027.