South Korean manufacturers will hand over previously undisclosed production data containing the know-how of skilled workers to a government research program on physical AI. Artificial intelligence trained on that data will control actual equipment at factories in Gyeongsangnam-do, then learn from the results to improve its performance. In Jeollabuk-do, AI will run a "dark factory" in which it designs and operates an entire plant by integrating robots and equipment from different makers. Technology validated in the two regions will be spread to other production sites and eventually developed into an export model in the form of "K-manufacturing plants."
The Ministry of Science and ICT unveiled the five-year research and development roadmap for 2026 to 2030 at a launch briefing for the Jeollabuk-do and Gyeongsangnam-do physical AI research programs, held on the 16th at the Novotel Ambassador in Seoul.
The government this year announced physical AI research programs worth a combined 1.4131 trillion won ($1.01 billion) for the two regions, with Gyeongsangnam-do focusing on precision manufacturing and Jeollabuk-do on factory platforms.
In Gyeongsangnam-do, the project will demonstrate physical AI specialized in precision control, with AI working alongside people on the factory floor and carrying out actual tasks. At its core is a large action model. Manufacturing data covering the movements of workers and robots, the laws of physics and spatial information will be used to train AI to understand production processes and control real equipment. Digital twins, data pipelines and infrastructure based on neural processing units will be linked to build a cycle running from data training to on-site application and then retraining on the results.
That requires data on the know-how skilled workers have built up on factory floors over many years. Cha Suk-won, a professor at Seoul National University, said 13 companies in Gyeongsangnam-do have agreed to share tacit-knowledge data they use in their own plants. "We will collect and process the data to develop physical models for each process, and the models we build will be packaged so that other manufacturers with similar processes can use them," the professor said.
In Jeollabuk-do, the project will upgrade a physical AI software platform tailored to manufacturing and logistics sites. The goal is "factory orchestration," in which different robots and equipment work together so the entire plant operates as a single system. AI will design the factory layout and the movement paths of logistics robots, verify them in advance using a three-dimensional digital twin, and then control the actual robots and equipment in an integrated manner. The aim is a general-purpose "dark factory" in which AI handles everything from design to construction and operation. In a proof-of-concept last year, AI completed in about three hours a robot factory design that had taken three to four experts nearly a month.
Chang Young-jae, a KAIST professor leading the Jeollabuk-do project, said the approach treats a factory as one giant robot made up of various robots. "Shinsung E&G, a supplier of semiconductor and display materials, parts and equipment, has decided to invest 20 billion won, and we will also build an ecosystem that links physical AI research to regional investment," the professor said.
The government has placed factory floors at the center of physical AI development to turn the process knowledge and data accumulated in Korean manufacturing over decades into a technology asset for the AI era. Until now, manufacturing AI has mainly been used to detect defects or anomalies from data, but the plan is for AI to learn the knowledge of experienced workers and apply it to running and controlling plants. "The factory floor is a living textbook where countless workers have poured in knowledge over decades of breathing alongside machines," Cha said. "The earlier approach was to look at data and find what was wrong, but now AI absorbs the knowledge embedded in the field, and that is the biggest difference in this project."
The government will invest a total of 1.4131 trillion won from 2026 to 2030, with 736.8 billion won for the Jeollabuk-do project and 676.3 billion won for the Gyeongsangnam-do project. Over the following five years, it will upgrade full-stack physical AI capabilities and spread them to manufacturing sites across the country. The ultimate goal is to develop an industrial ecosystem capable of entering overseas markets in the form of "K-manufacturing plants."
