A robot lifts standardized boxes from a conveyor belt. Another adjusts the position of barcode-labeled boxes for scanner alignment. This represents the current level of robotic technology deployed in automotive and other manufacturing facilities.
South Korea ranks among the world's leaders in manufacturing robot installations. However, robots are not deployed across all production processes. They still cannot handle precision tasks requiring fine motor skills—like threading a needle—or non-standardized work.
Experts predict that advances in artificial intelligence and physical AI could enable robots to perform practical manufacturing tasks within two years. If supporting technologies develop accordingly, robots could replace the shrinking working-age population in an era of low birth rates and aging demographics, analysts say.
The evolution of robots from "machines" to "workers" ultimately depends on how quickly they can accumulate and replicate human expertise as data.
Data Learning Is Key to Robot Deployment
For robots to perform various tasks in automotive, shipbuilding, defense, and steel manufacturing, they must first acquire and learn relevant operational data.
"Korea has largely established automated processes with substantial accumulated data, and can secure even more going forward," said Yoon Seok-jun, head of POSCO DX's Robot Automation Center. "However, whether we have sufficient data for robot learning is another matter entirely."
Systems must be built to identify and secure the data necessary for robot learning, he noted.
Industry experts identify "imitation learning" as the fundamental method for data acquisition and robot training. Operators control devices like remote controllers while robots mirror their movements, continuously collecting motion data. AI can then significantly expand this learned dataset. More data enables robots to move with greater accuracy and precision.
World foundation models represent another key learning method. These create virtual replicas of actual work environments where robots train through simulation. Repeated practice in virtual spaces gives robots predictive capabilities for various scenarios, enabling responses to unexpected situations. However, minimizing the gap between simulation and reality remains a challenge.
Additional methods include extracting data from worker video footage or attaching sensors to workers' bodies to capture movement data for robot training. Pre-trained robots then undergo fine-tuning to adapt to specific manufacturing environments.
Even after this process, robots cannot immediately deploy to production floors. Reality proves more difficult than theory—much like having excellent textbooks doesn't guarantee good grades. Robots also experience "catastrophic forgetting."
"A robot that learned 10 things might learn 2 more, then experience conflicts between old and new learning that cause it to forget what it knew," explained Park Sang-yeop, Chief Technology Officer at LG CNS. "Even one mistake during training can drastically reduce accuracy."
While securing abundant training data is important, data curation—accurately distinguishing which data will aid robot training—proves equally critical.
Finger and Joint Counts Vary by Task
Hyundai Motor Group's "Atlas" robot, unveiled at CES 2025 in the United States last month, has three fingers. LG Electronics' towel-folding robot "CLOi" displayed at the same event has five fingers. Finger count varies based on deployment site and task characteristics. Simple gripping requires only two fingers, but complex tasks demand more.
Joint count also matters. Robot manufacturers differ in their designs, with hands typically having 22-23 joints. Humanoid robots have approximately 80 total joints. Depending on type and purpose, robots may have 30-40 joints or over 100. More joints enable more human-like movement.
Transferring Craftsmen's Expertise
Proper data training and task-appropriate hardware don't guarantee immediate commercial deployment. Worker know-how remains a key challenge. No amount of training data can easily replicate expertise workers acquire over years on the job.
For robots replacing humans in high-temperature environments, preventing thermal deformation is essential. When multiple robots work together, collision-avoidance systems are mandatory. Systems predicting equipment failures and optimizing operations based on battery charging times are also necessary. New factory designs must incorporate all these considerations, driving growing demand for related consulting services alongside robot commercialization.
Various automation technologies currently operate in manufacturing. Demand is highest for dangerous, simple tasks. POSCO's steel mills, handling high-temperature and high-pressure equipment, use automation for removing dross (contaminants) from product surfaces, inserting temperature sensors into molten steel at thousands of degrees, and cutting bands from coiled steel plates.
As technology advances, "robot employees" could be hired for highly complex tasks. "Battery work alone involves high-voltage electricity with electrocution risks," Park said. "Chemical processes require workers to wear masks and protective suits. Robots could dramatically reduce such inconveniences."
An Optimized Environment for Industrial Robots
Robot-related IT companies assess South Korea as optimally positioned for industrial robots. The country has the world's highest robot density relative to population. According to the International Federation of Robotics, Korea has 1,012 robots per 10,000 workers. Manufacturing sectors including semiconductors, shipbuilding, nuclear power, and secondary batteries are well-developed, with data collection and automation systems established across various processes.
This explains why global startups developing Robot Foundation Models—the "brains" of robots—actively pursue partnerships with Korean companies. POSCO Group, with its manufacturing expertise, is also developing its own RFM.
Industrial robot prices currently range from several million won to hundreds of millions of won depending on manufacturer. Companies purchasing multiple robots need returns exceeding their investment. Fortunately, technological advances are driving prices down. Unlike humans, robots can operate 24 hours a day, 365 days a year, leading analysts to favor positive return-on-investment projections.
"Maintenance is as important as installation," Yoon said. "Minimizing deployment time through virtual commissioning when processes change is economically critical."
Industry executives agree that robots in manufacturing signify more than simple labor replacement. As working-age populations decline rapidly due to aging demographics, robots can address labor shortage issues.
"Going forward, humans will make critical decisions in control rooms, AI operators will issue commands, and robots will execute them in 'intelligent factories,'" Yoon said. "The working population will decline significantly within 10 years. Companies without intelligent factories will inevitably face competitiveness problems."
Additionally, robots inheriting the capabilities of field experts and master craftsmen could create new revenue models. Ryu Jung-hee