LG Electronics and Nvidia have agreed to secure 100,000 hours’ worth of robot training data within this year. That is equivalent to a single robot working nonstop for 12 years.
On the 18th, LG Electronics invited key Nvidia officials to the “LG Data Factory” under construction at the former Yangjae R&D Campus in Seocho-gu, Seoul, to discuss cooperation in robotics and ways to commercialize the business.
LG Electronics CEO Reo Je-cheol, LG CNS CEO Hyun Shin-gyun, and LG Sciencepark CEO Jeong Su-heon were among the group’s top executives who attended. On the Nvidia side, Senior Director Madison Huang and other key officials visited and toured the data augmentation and synthesis facilities.
The meeting follows a future strategic business memorandum of understanding signed on the 13th at Nvidia’s headquarters in Santa Clara, California. With the two companies meeting again at a data production site just five days after the agreement, observers say the partnership has moved beyond the declaration stage and into commercialization.
◆ Robots have no textbooks
Artificial intelligence such as ChatGPT learned from texts and images accumulated on the internet as its textbooks. Robots are different. There is no movement data on the internet for tasks such as picking up and moving objects.
To obtain data, robots must be made to work in real-world environments. That is why the competition for humanoid robots is shifting from hardware performance to the securing of training data.
AI that moves its body in physical space is referred to in the industry as “physical AI.” It is a technology that must understand object weight, distance, and even force control. Nvidia, the No. 1 AI semiconductors company, is leading the market in this field by promoting development tools that serve as virtual training grounds.
LG Electronics’ strength in the partnership lies in the real-world work data accumulated over decades in home appliance factories and logistics sites. The company plans to train robots by adding the know-how of skilled workers and then greatly expand the data using Nvidia’s technology.
◆ Building 12 years of work data in one year
The Yangjae Data Factory is being built with a total floor area of 10,000 square meters across one basement level and three floors above ground. LG Electronics’ in-house domestic robot, “LG Cloyd,” has already been deployed there to generate data.
Cloyd repeats tasks such as cleaning in a space designed to resemble a home, as well as moving and assembling parts in a manufacturing environment modeled after LG’s Tennessee washing machine plant. Robots have also been deployed in LG CNS’s logistics automation facility and in LG Innotek’s robot hand training space.
Data generated by robots working in real settings is then expanded through Nvidia’s simulation technology. The method is to create countless similar work situations in virtual space inside a computer to increase the amount of data. It is similar to a driving trainee gaining experience through simulator training in addition to road practice. This technology compensates for the limits of the robotics industry, where gathering unlimited amounts of real data is difficult, by using virtual data.
LG Electronics plans to increase the number of deployed robots into the hundreds by year-end and secure a total of 100,000 hours of training data by combining real-world collections with virtually generated data. The goal is a virtuous cycle in which robots improve after learning from the data, and improved robots then generate even better data. The data accumulated this way will be used to enhance the performance of the robot foundation model, or RFM, which serves as the robot’s brain.
◆ Selling everything from components to the brain
LG Electronics has designated this year as the first year of a leap forward in its robotics business. Last month, it established the Robotics Business Center directly under the CEO to oversee the company’s robotics business. It is expanding its product lineup from industrial and commercial robots to household robots, while also strengthening capabilities in developing core components such as actuators, which drive robotic joints.
The goal is to become a “robotics total solution provider” covering components, finished products, training, and operations. In other words, the company aims to supply not only robots but also the data and software needed to train them.
CEO Reo Je-cheol said, “Through synergy based on the ‘One LG’ principle, which brings together the group’s core capabilities, and strategic cooperation with global partners, we will secure competitiveness in physical AI and become a robotics total solution provider.”
For manufacturers that directly own factories and logistics sites, the humanoid competition is like having a production base for data as well. Analysts say this strategy, which turns one’s own operations into a data asset rather than remaining focused only on hardware competition, could serve as a reference model for the broader domestic manufacturing sector.