The Gyeonggi Research Institute has proposed that Gyeonggi Province build an ecosystem centered on industrial complexes that links data, demonstration facilities, corporate support, and talent development in order to expand the adoption of physical artificial intelligence (AI) in manufacturing.
The institute published a report on the 6th titled "The Era of Physical AI: The Reality of Manufacturing and Gyeonggi Province's Response Strategy" containing these findings.
Physical AI refers to technology that goes beyond generating text or images like generative AI, combining with physical devices such as robots, machines, and automobiles to perceive and judge reality and carry out actual tasks. In manufacturing settings, it is used for product defect inspection, equipment failure prediction, robot task control, and process optimization.
According to a survey of 200 manufacturing firms that had reached at least the basic stage of smart factory adoption, 77.5% had pursued digital transformation across their manufacturing processes. Some 92.9% had built manufacturing execution systems (MES), and 79.4% had introduced enterprise resource planning (ERP) systems. Digitalization at the level of individual systems has thus advanced considerably.
However, only 12.5% of firms were using physical AI, in which AI is combined with production equipment for tasks such as inspection, movement, process management, and optimization control. Among current applications, recognition functions that identify objects or products were the most common at 96.8%. Prediction functions stood at 56.8% and automation functions at 36.8%. That said, 37.0% of firms said they would newly introduce or expand physical AI within the next three years, confirming the potential for wider adoption.
The level of data utilization was even lower. While 44.0% of firms managed production data by linking it with MES and ERP, only 1.5% used it for AI learning and process optimization. Just 2.0% of firms had the high-performance computing environment needed for AI learning and simulation.
The main reasons firms are considering adopting physical AI were quality improvement at 74.0% and productivity improvement at 73.5%. On the other hand, the burden of initial investment costs at 69.7% and the burden of replacing existing equipment at 50.3% were cited as major obstacles. When making investment decisions, the reliability and stability of the technology at 49.0% and compatibility with existing equipment at 35.0% emerged as important conditions. Some 41.5% of firms expected to recover their investment within two to three years.
In terms of securing personnel, retraining incumbent employees was presented as more central than new hiring. Some 79.0% of surveyed firms preferred training existing workers and reassigning them to new tasks. As for the effects of adopting physical AI, assisting skilled workers and easing labor intensity was the highest at 44.5%.
The report stated that Gyeonggi Province should standardize manufacturing data by industry, including semiconductors, automobiles, and machinery, and build demonstration testbeds and manufacturing data platforms in industrial complexes. It proposed that verified technologies be spread to firms within the province by linking them with vouchers, consulting, testing and certification, and investment support.
"The key to spreading physical AI is not introducing equipment but connecting data, production facilities, supplier companies, and on-site workers," said Bae Young-im, senior research fellow at the Gyeonggi Research Institute. "Gyeonggi Province must establish an integrated support system that covers everything from identifying firms' on-site problems to demonstration, commercialization, fostering specialized companies, and strengthening the capabilities of incumbent workers."
