SKT to Apply "Physical AI" to Manufacturing Sites... Aiming to Make It a Core Pillar of AI Business

"Validating Processes in Virtual Environments to Drive Manufacturing Innovation"

Jaeheon Jung, CEO of SK Telecom.

On the 11th, SK Telecom announced that it is introducing "Physical AI," which combines digital twin technology and robot training, to address difficulties in applying artificial intelligence (AI) in manufacturing settings due to complexity and a lack of verification environments. Through this approach, the company aims to validate processes and train robots in virtual environments, accelerating manufacturing innovation and planning to expand into various industries.

The digital twin platform is built on NVIDIA's Omniverse-based technology. SK Telecom is integrating features required by industry and evolving the technology into a form that can be applied to real manufacturing processes. Notably, data generated from different facilities and systems can be reflected in a single virtual environment, and simulations that adjust various conditions allow for scenario-based pre-verification of operations before process changes or new equipment introductions. This reduces trial and error and supports more stable decision-making.

NVIDIA Omniverse is a platform that enables simulations in a precisely rendered 3D virtual space modeled after real-world environments. It incorporates the laws of physics to realistically replicate the movements of equipment, processes, and robots, and allows for the integration and analysis of diverse data within one environment. This makes it possible to pre-validate a variety of scenarios before implementing process changes.

The robot training platform utilizes the digital twin platform's virtual environment to train robots. To address the challenge that direct robot training on actual manufacturing lines is difficult, the system enables repeated validation and learning under various work conditions and situations in the virtual environment. In this way, the robot's responsiveness and stability can be secured before deployment to real-world sites. In this process, synthetic data that closely resembles real-world data is artificially generated and used to supplement the lack of actual data, and robots are trained through the Robot Foundation Model (RFM) learning to perform a variety of tasks.

Baek Seungyoon, Head of SK Telecom's Data Intelligence team, stated, "We see Physical AI not simply as an automation technology, but as a core infrastructure for shifting the operational paradigm of manufacturing toward AI-centric processes," adding, "We plan to expand its application to various manufacturing industries, including semiconductors, automobiles, and shipbuilding, and focus on creating real value for AI in the manufacturing field."

This content was produced with the assistance of AI translation services.