South Korea has developed a "manufacturing-specialized" agentic artificial intelligence (AI) that understands human language and independently makes decisions and takes actions. The technology is being hailed as a potential lifeline for resolving chronic skilled worker shortages in regional manufacturing industries while shortening process times.
The Korea Electrotechnology Research Institute (KERI) announced Monday that a team led by Dr. Lee Joo-kyung at its AI Research Center, in collaboration with Changwon National University, has developed an "autonomous manufacturing multi-agent AI" technology that enables multiple AI robots to collaborate and actively operate processes.
Conventional factory automation robots have been passive machines that move only according to code pre-programmed by experts. As a result, whenever the working environment changed even slightly or new components were introduced, responsible engineers had to stay up for days rewriting code. This placed enormous time and cost burdens on regional small and medium-sized enterprises (SMEs) facing workforce shortages.
As a solution, KERI developed an agentic AI technology based on a large language model (LLM) that understands commands and independently establishes optimal work plans.
At the core of the technology is intelligent task division through "multi-agents." Just as a foreman issues instructions, a language agent delivers commands, while vision and robot control agents communicate with each other to divide roles.
The research team also dramatically overcame the limitations of "grounding," or real-world perception, technology in this process. Previously, when instructed to pick up "that red part over there," robots often failed to accurately understand everyday referential expressions such as "over there" or "the red part."
KERI improved the language agent's ability to grasp work intentions, after which the vision agent analyzes the precise three-dimensional coordinates of objects through cameras to generate control scenarios, and the robot agent operates precisely without error based on the scenarios it receives. This completes what is known as "Actionable AI," in which intelligence in the virtual world seamlessly translates into robot movements on the actual site.
Organic collaboration is achieved with a single simple command, eliminating complex coding procedures. As a result, process reset work that previously took nearly a week is expected to be reduced to within one hour. "The technology instantly adapts to objects or environments it encounters for the first time and can reduce additional software modification costs associated with process changes, making it ideal for SMEs that need small-batch, high-variety production," the research team emphasized. Dr. Lee added, "Regional small and mid-sized companies have been hesitant to adopt AI due to high costs and workforce shortages. Our technology is a solution that can smartly transform existing manufacturing lines without significant costs, and we will contribute to enhancing the competitiveness of regional manufacturing through technology transfer."
The achievement is particularly significant as a breakthrough in the "VLA (Vision, Language, Action)" field, where global big tech companies such as Google, Nvidia, and Tesla are competing. "Overseas technologies are centered on large models, making them heavy to operate and dependent on specific hardware, whereas the technology developed this time has been lightened and modularized to suit manufacturing sites, securing world-class competitiveness in terms of 'on-site applicability' that can be immediately deployed in actual factories," KERI said.
Going forward, KERI and Changwon National University plan to nurture the students who participated in this convergence research into "advanced AI talent" that can be immediately deployed to regional industries, building a virtuous ecosystem that addresses the national challenges of population decline and declining manufacturing competitiveness.
