• Deputy Prime Minister and Minister Bae visited Shinsung Deltatech in Gyeongnam to review the results of the pre-verification project for physical AI and meet with local manufacturers.
• MSIT is set to embark on the large-scale “Gyeongnam AX” project in the first half of 2026 to transform manufacturing sites with physical AI.

【Relevant National Tasks】 

• 21. Becoming the world’s most AI-literate country
• 22. Securing “unmatched” frontier technologies and talent in artificial intelligence (AI).

Deputy Prime Minister and Minister of Science and ICT (MSIT) Bae Kyung-hoon visited Shinsung Deltatech in Changwon, Gyeongnam Province on Friday, February 6 to review the results of the pre-verification project for physical AI and to discuss ways to improve regional manufacturing competitiveness.

In addition to reviewing the project results, this visit was designed to confirm the applicability of precision control technology based on physical AI and to hear directly from companies and researchers on how to connect this technology with the upcoming large-scale research and development (R&D) project called “Gyeongnam AX (AI Transformation).”

The Gyeongnam AX project, which will be launched by MSIT this year, aims to incorporate the physical characteristics of manufacturing environments and the know-how of skilled workers into AI models, rather than simply automating production processes. To achieve this goal, MSIT is focusing on developing PINN (Physics-Informed Neural Network)*-based LAM (Large Action Model)** as this technology enables AI to control robots and facilities directly. The successful development of PINN-based LAM technology means a shift from existing AI focused on analysis and judgement to AI with deep knowledge of industrial sites that can actually control manufacturing processes.

* PINN (Physics-Informed Neural Network): An AI model that incorporates a variety of physical laws that may be encountered in the course of manufacturing processes.

** LAM (Large Action Model): A next-generation AI that actually performs action, rather than simply functioning as an LLM.

▲ Physical AI Pre-verification Project (Non-R&D)

※ Responsible Agency: The National IT Industry Promotion Agency (NIPA)

• (Period and Budget) August – December 2025 (5 months) / KRW 19.7 billion
• (Participants) Gyeongnam University (Lead), Seoul National University, and eight manufacturers based in Gyeongnam Industrial Complex *

* Shinsung Deltatech, Hwaseung R&A, CTR, Korens, KG Mobility, GMB Korea, Samhyun etc.

▲ The Gyeongnam AX Project for Physical AI (R&D to begin in the first half of 2026)

• (Period and Budget) 2026 – 2030 / Project scale to be determined after reviewing its appropriateness.
• (Details) Developing PINN-based LAM technology to implement physical AI for precise control.

With the 2025 supplementary budget, MSIT and the National IT Industry Promotion Agency (NIPA, President: Park Yoon-gyu) have been carrying out the pre-verification project and conducted an on-site review at eight manufacturers in Gyeongnam Province.

This verification revealed tangible results among participating companies in the forecasting of manufacturing process quality and the enhancement of production efficiency. For instance, Shinsung Deltatech built an AI training dataset linking its 49 plastic injection molding process data points and 62 action data points (including worker behavior, raw material conditions, and defect shapes) to the plastic injection and assembly process. Subsequently, the company used a digital twin model to predict and proactively adjust process quality. The result was a potential reduction in defect rate of approximately 15% and an increase in facility utilization rate of approximately 20%.

On the other hand, Hwaseung R&A predicted potential material deformation that may occur during the rubber extrusion process, thereby improving overall facility efficiency by at least 5%. As for CTR, they predicted chattering, a machine vibration that occurs during aluminum machining processes, reducing the defect rate and shortening the machining cycle time by more than 17%.

Based on the results of the pre-verification, MSIT is set to embark on the Gyeongnam AX project in the first half of 2026. The project, which is scheduled to run through 2030, will focus on developing a physical intelligence behavior model to implement physical AI for precise control.

At the following session, representatives from participating companies discussed topics such as the expansion of physical AI-based precision control technology, strategies for its implementation, data management, and the conversion of workers’ know-how into AI models. They also called on the government to strengthen policy support.

Gyeongnam has favorable conditions for the accumulation and verification of field-oriented data as it is a manufacturing hub of advanced industries. MSIT plans to actively reflect the feedback gathered during this visit in its AI transformation strategies and support policies in the future to ensure that the results can be translated into practical use.

“Gyeongnam offers an ideal environment full of companies with precision manufacturing capabilities in equipment, parts, and machinery,” said Deputy Prime Minister and Minister Bae. “By integrating physical AI with the province’s industrial competitiveness, the government will make it a strategic hub of the “5-Mega, 3-Special” regions.”

He added that physical AI is a key technology that will determine the future of Korea’s manufacturing competitiveness, stressing that systematically building data generated at manufacturing sites is all the more important at this stage.

He also said that once a physical AI foundation model is developed based on this data, the government will distribute it by making it open source to foster an ecosystem in which many companies can easily access and leverage the technology. “Rather than being satisfied with short-term success, the government will continue to carry out physical AI policies with a strong sense of responsibility for Korea’s future,” he added.