[fn Person] “Before Physical AI Is Deployed in the Field, We Train It in a Virtual Environment”

[Financial News]

• Lee Jae-cheol, CEO of SKAI Intelligence
• Physical AI commercialization hinges on ‘experience’ in real-world environments
• Data alone is not enough to provide sufficient training
• Deliberately creating failure scenarios in virtual environments
• Collaborating with ABB Robotics and Seoul National University

“The key bottleneck to the commercialization of Physical AI is not robot hardware, but the lack of data and experience needed to safely and repeatedly learn from and validate the wide range of situations encountered in the real world.”

Lee Jae-cheol, CEO of SKAI Intelligence (pictured), identified data as a key factor that will determine the future of the Physical AI race. While hardware and AI models—the equivalent of a robot’s body, eyes and brain—are advancing rapidly, data and experience required for robots to operate reliably in complex real-world environments remain insufficient, he explained.

SKAI Intelligence is a digital-twin-based Physical AI data infrastructure company that precisely virtualizes real-world environments to support the training and validation of AI and robots. Lee described the company as “a company that creates training materials and test environments so that robots can experience a wide range of situations in a virtual environment before being deployed in the field.”

The company started out in 3D content production, virtually recreating products and physical spaces. After participating in industrial Physical AI projects with ABB Robotics as an NVIDIA ISV partner, the company identified an opportunity to expand its capabilities into synthetic data for robot training and validation.

Synthetic data is essential because real-world data alone cannot provide sufficient training and validation for robots. “Real-world data takes a great deal of time and money to collect, and it is difficult to capture every possible exception,” Lee said. “This is particularly true in precision manufacturing, where data can often only be collected after a product and production line have been completed.”

Lee compared the situation to a student who gets all the answers right on practice tests but encounters an unfamiliar variation of a problem for the first time in the real exam.

Synthetic data addresses this limitation by deliberately creating variations and failure scenarios in virtual environments. Lee described the technology as “a process of transferring reality into a virtual environment, generating countless scenarios to validate AI and robots, and then feeding the results from the real world back into the virtual environment.” He added, “We turn robots’ failures into an answer sheet of wrong answers and use them to reduce the gap between the virtual and real worlds.”

“The differentiating factor of SKAI Intelligence is that we have connected engineering data from industrial sites with high-precision CG and simulation technologies into a single pipeline,” Lee emphasized. “We apply technologies accumulated through our content business to industrial data generation, and have established a closed-loop system that connects digital-twin construction, synthetic data generation, robot validation and field feedback.”

SKAI Intelligence is advancing its technology based on NVIDIA Omniverse and Isaac Sim. In June, the company signed a cooperation framework agreement with ABB Robotics and began a proof-of-concept project linking its synthetic data pipeline with robot simulation environments. It is also conducting joint robotics research with the AI Research Center at Seoul National University.

Lee expects the next five years to be a period in which the formats, quality and validation standards for Physical AI data will be established. “At present, each company and platform uses different methods of generating and representing data, making it difficult to use data created in one environment in another,” he said. “We will work with leading companies and research institutions to create real-world results and participate in the process of establishing industry standards.”

wongood@fnnews.com — Joo Won-kyu, Staff Reporter