NVIDIA has emphasized the importance of simulation technology to expand the deployment of physical artificial intelligence (AI) in industrial settings. The company explained that for robots and autonomous vehicles to operate safely in real-world environments, they must secure sufficient data in virtual environments and reduce errors that occur when applying that data to reality.

Chung So-young, Country Manager of NVIDIA Korea, delivered a special lecture titled "Physical AI for Next-Generation Industrial Digitalization" on the 13th at the "Physical AI Tech Week" event held at COEX in Gangnam-gu, Seoul. She emphasized the necessity of learning through simulation, stating, "In real-world environments, data is inevitably scarce, and issues regarding safety and cost arise during the process of deploying AI."

Chung presented the "Sim-to-Real Gap" as a major challenge in physical AI development. This refers to a phenomenon where robots trained in a virtual environment do not operate as expected in the real world.

She explained, "No matter how realistic simulation data is made, it may not operate accurately because it differs from the actual environment." This is because it is difficult to perfectly implement the physical laws and various variables of the real world into a virtual environment. To solve this, she explained that a combination of additional learning using real-world data and simulation data is necessary.

To reduce these limitations, NVIDIA provides physical AI-dedicated simulation and data generation technologies. Representative technologies include the 3D simulation platform "Omniverse" and the world foundation model "Cosmos."

Omniverse is utilized to implement virtual environments similar to actual industrial sites, while Cosmos is an AI model that generates learning data reflecting physical environments and the movement of objects.

Chung also emphasized the need for computing platforms and robot foundation models required for physical AI development. NVIDIA's "Isaac GR00T" is a model that trains robots to perceive their surrounding environment and perform various actions, whereas the robotics computing platform "Jetson" plays the role of running the trained AI models when mounted on actual robots.

Chung stressed that because physical AI development requires large-scale 3D data processing and simulation, the computing infrastructure to support it is equally important.

She also evaluated South Korea's industrial competitiveness in physical AI positively. She forecasted that South Korea can significantly contribute to the development of related industries, stating, "South Korea has a strong manufacturing foundation and the government is also showing a lot of interest in physical AI."

[ⓒ Global Economic Times. 무단전재-재배포 금지]