Meet the robot dog patrolling the hottest spot in Posco's mill

A robot dog spends its days inside a $7,000 “home” of its own at Posco’s steel mill in Pohang, North Gyeongsang. The house has an automated door that shields the interior from clouds of metallic dust as well as an air conditioner to cool the chamber.

As soon as the robot, called Spot, receives a signal from one of its handlers, Hwang Seok-gyun, a senior researcher at Posco’s research institute, it heads straight to work without a second’s hesitation.

Its primary responsibility is monitoring blast furnaces filled with molten metal exceeding 1,500 degrees Celsius (2,732 degrees Fahrenheit). Despite being encased in ceramic walls 3 meters (10 feet) thick, microscopic cracks can release high-pressure and superheated gas, posing a risk to human workers.

To mitigate that danger, the robot dog patrols the site using thermal imaging and high-resolution cameras mounted on its body, inspecting 30 oxygen injection ports and hundreds of interconnected pipes for abnormalities.

At the top of every hour, it rises and moves first toward the oxygen injection ports, fixing its gaze on them for 10 seconds before turning to inspect pipes and valves for another 10 seconds. If it detects any irregularities such as temperatures straying beyond the normal operating range of 20 to 25 degrees Celsius, it immediately alerts Hwang.

It is now just another ordinary day inside a Posco steel mill. Before the robot dog was deployed at the plant in October 2024, the task required at least two production workers rotating around the clock.

The robot’s workday never truly ends, repeating a relentless cycle of 40 minutes of inspection followed by 20 minutes of "rest" inside its kennel. While the robot dog patrols the factory floor, a handler monitors and analyzes the incoming data from the comfort of a control room one floor below.

“People should never get too close to the blast furnaces, as at any moment, someone could get seriously hurt,” said Hwang. “Now the robot dog goes into those dangerous areas in place of workers, and as its handler, I feel both proud and grateful.”

“I hope humanoid robots will soon be introduced as well, taking on tasks Spot still cannot perform, such as manually shutting valves,” Hwang added.

The robot dog and priced at roughly 300 million won ($200,000), is made by Boston Dynamics, a U.S. robot subsidiary of Hyundai Motor. Posco purchased it in 2023, and a total of three robot dogs are deployed across Posco’s steelmaking facilities.

Robots at heavy-material scenes

Concerns were especially acute around the blast furnaces, where summer temperatures can soar beyond 40 degrees Celsius. Although Spot was engineered to withstand environments ranging from minus 20 to 50 degrees Celsius, the handlers remained uneasy about exposing such an expensive machine to the punishing heat of the steel mill.

To prevent mishaps, Posco commissioned a nearby industrial workshop to build a high-tech kennel tailored specifically to the robot. The shelter and factory floor were equipped with light-emitting diode lighting and surveillance cameras, enabling Spot to safely navigate even during overnight operations.

Posco plans to buy more robotic dogs in the future and has been working with domestic robotics manufacturers — including Rainbow Robotics, LionsBot and AIDIN Robotics — to develop robot dogs better suited to the operational realities of steel mills.

“When we develop robots together with local companies, it becomes much easier to customize them for our production environment and repair them quickly if something breaks down,” Hwang said.

Robots taking on dangerous industrial tasks are rapidly becoming a familiar sight across Korea’s traditional manufacturing sectors such as steelmaking and shipbuilding.

Hyundai Steel has introduced robotic systems to unload massive steel coils, while HD Hyundai Heavy Industries (HHI) has deployed wall-climbing robots capable of inspecting welds and cleaning the towering outer hulls and storage tanks of ships.

Hanwha Ocean uses drones to inspect equipment installed atop enormous shipbuilding blocks and other elevated structures.

Robots as safety net, not job threat

“Robots take away human jobs? That sounds like something people say when they don’t really understand how our industry works,” said Oh Jung-cheol, a master technician who has spent 39 years at HD Hyundai Heavy Industries, adding that the growing presence of robots inside shipyards is not a threat but a welcome change.

From overseeing the installation of massive ship engines to supervising the assembly operations of the prefabricated modular section of a ship's hull at the company’s shipyard in Ulsan, Oh has watched robots steadily expand into fabrication, welding and painting operations over the past decade.

“When shipyard workers are pushed through continuous high-intensity labor, the risk of accidents such as getting caught in equipment, falling from heights and other serious incidents rises sharply,” Oh said. “By deploying robots in those tasks, we’ve dramatically reduced the number of workers sent into hazardous environments. I see robots as a protective shield against industrial accidents.”

One steelworker at Posco’s steel mill agreed, saying workers on the factory floor have “largely welcomed the introduction of robotics.”

Having spent more than a decade beside blistering blast furnaces, the veteran worker said he could already sense the looming impact of what many in the industry now call the “Atlas shock” — a reference to the rapid rise of humanoid robotics. Yet rather than resisting the shift, he described it as an irreversible transformation already underway.

“The fear that robots might someday replace my job feels abstract compared to the very real dangers we face every day,” he said. “Right now, safety matters far more.”

Robots on the frontline

Until recently, steel mills and shipyards had been relatively slow to phase in robotics. Aging infrastructure and production lines heavily dependent on complex manual labor meant that only small collaborative robots were selectively deployed for particularly hazardous tasks.

That contrasts with other manufacturing sectors such as automobiles, where robots already handle more than 90 percent of repetitive processes such as welding and painting, while semiconductor and display factories operate near-fully automated systems built to tolerate not even a speck of dust.

That calculus began to change rapidly after Korea’s Serious Accidents Punishment Act came into full force in January 2024, ending a grace period on enforcement and placing greater legal responsibility on companies — and individual leaders — to prevent industrial accidents.

That challenge has coincided with the structural problem of manufacturers — the aging of highly skilled workers. Across the industrial sector, the consensus on factory floors is increasingly clear that there is simply no alternative as viable as robots.

China had the world’s largest installed base of industrial robots in 2024, with 2.03 million units operating in factories, followed by Japan with 450,500 and the United States with 393,700, according to data from the International Federation of Robotics. Korea ranked fourth globally with 391,900 industrial robots in operation.

Yet when measured by “robot density” — the number of bots per one million workers — Korea far surpasses even China. Despite China’s population of 1.4 billion, its robot density stood at 1,436, compared to Korea’s world-leading 7,595.

“For manufacturers under mounting pressure to improve productivity, robots are no longer a choice but a necessity,” said Lee Jae-yeol, head of the Integrated Robot Systems R&D Division at the Korea Institute of Robotics and Technology Convergence, which jointly develops robots with private companies including Posco and HD Hyundai.

Robots over master craftsmen

In highly skilled trades such as welding, it can take a decade to cultivate a true master. That can now be absorbed by robots almost instantly through machine learning.

Welding is often called “the heart of the shipbuilding” as nearly 60 to 70 percent of the ship construction process involves welding. Yet earning a reputation as a genuinely skilled welder typically takes at least 10 years on the factory floor.

As ships must endure decades of stress from rolling waves and the immense weight of tens of thousands of tons of cargo, airtight precision is the essence of welding. Even a single air bubble within a welded seam is considered unacceptable.

Unlike automobiles or home appliances, where thin metal sheets are stamped and assembled mechanically, the thick steel plates used in shipbuilding require experienced workers to manually hammer, align and join each section by hand with painstaking precision.

An exodus of these experienced workers, however, has left a gap. Assembling a single ship block typically requires 50 to 60 workers, with welding alone accounting for roughly half of the labor. In reality, though, many shipyards now have only 10 to 15 welders available on site.

And training replacements isn’t simple. Welding is a craft that compresses a decade’s worth of experience and tacit knowledge into every movement, making it difficult to rapidly train foreign contract workers who frequently rotate in and out of shipyards.

“Many highly skilled workers painstakingly trained by shipbuilders are increasingly leaving for freelance subcontracting crews that offer higher daily wages,” said Yang Seung-hoon, a sociology professor at Kyungnam University. “With shipbuilders recently flooded with new orders, the shortage of skilled labor at worksites has reached a critical point.”

So companies are turning to robots for solutions. HD HHI operates 133 robots, while HD Hyundai Samho had deployed 90 as of the end of February. Hanwha Ocean runs around 80 welding and collaborative robots across its shipyards.

HD HHI plans to introduce an additional 30 welding robots developed by Rainbow Robotics, while Hanwha Ocean is conducting in-house research and development on advanced welding robots through its engineering arm.

“Once new welding robots are developed, they are being deployed to production sites almost immediately,” a spokesperson for Hanwha Ocean said. “The technology is also expected to be introduced at Hanwha Philly Shipyard in the United States, where shortages of skilled welders are particularly severe.”

Can robots replace veteran craftsmen?

Over the course of an entire workday, robots have already begun to outperform humans.

Measured task by task, human welders still hold the edge, as an experienced welder can complete two “cells” — a shipbuilding work unit equivalent to one-60th of a ship block — in roughly 13 minutes, compared to about 15 minutes for a welding robot, according to HD Hyundai Samho.

But that equation shifts dramatically in terms of daily work. A skilled worker typically completes 25 to 30 cells per day, while a robot can process 45 to 50 without interruption.

“Human productivity is inevitably affected by fatigue and physical condition,” a spokesperson for HD Hyundai Samho said. “Robots, by contrast, can maintain the same level of precision, speed and concentration throughout the entire day without deterioration in performance.”

If a human worker first inputs key parameters into a tablet, including the location of the material to be welded and the thickness of the weld line, robots then analyze the characteristics of the metal surface using touch sensors and identify the precise target line where the torch must make contact.

Once welding begins, the robot’s AI continuously recalibrates the machine’s position in real time, guiding the torch along the target line with microscopic precision.

The integration of AI has enabled welding robots to achieve precision welding within a margin of error of just 0.5 to 1 millimeter (0.02 to 0.04 inches), according to HD Hyundai Robotics.

But some tasks remain difficult to entrust to robots, such as curved-surface welding. Maintaining a constant distance and angle between the welding torch and the metal surface is critical, and robots still struggle to adjust in real time to the curvature of the hull. Even when robots perform real-time scanning with laser sensors, deviations still occur.

Curved welding also presents additional complications: molten metal can drip under gravity, while intense heat may warp the steel plates themselves. In such environments, the ability to respond instantly to subtle changes remains essential — an area where human judgment still outperforms machines.

For these highly sophisticated tasks, manufacturers have recently begun training welding systems through “imitation learning.”

Using cameras and high-precision sensors, engineers capture the movements of veteran welders as they work across curved surfaces, extracting detailed data points such as wrist angles, the distance between the torch and the metal surface, and the speed of hand movements.

Those datasets are then fed into AI models, which are trained not merely to replicate motions mechanically but to infer the reasoning behind them. Once embedded into robotic systems, the AI can apply those learned instincts to entirely unfamiliar steel plates with different curvatures, calculating optimal welding paths autonomously rather than relying on preprogrammed routines.

“The ultimate goal of imitation learning is for robots to internalize the instincts of skilled workers and independently determine optimal paths without humans manually inputting every piece of data,” said Lee Dong-jun, a mechanical engineering professor at Seoul National University. “If robots can inherit human expertise through what we might call a ‘digital succession of skills,’ it could help fill the growing vacuum of skilled labor in industries such as shipbuilding.”

Humanoids rush into mills and shipyards

Experts say beyond robot dogs and collaborative robotic arms, humanoid robots could eventually take over portions of complicated manufacturing processes such as welding.

“Conventional robotic arms and welding robots require dedicated rails, platforms and wide operating spaces to be installed throughout the factory,” an executive from a mid-sized shipbuilding firm said.

“Humanoid robots are different. Because they possess human-like proportions and can walk on two legs, they can move through existing worksites without major infrastructure changes,” he added. “They can step over thresholds and stairs on their own and navigate the complex interiors of ship structures much more freely.”

Companies have already begun moving from theory to deployment. Posco recently partnered with Persona AI to begin pilot-testing humanoid robots in port cargo-handling operations at Pohang later this year.

The humanoid will assist with loading and unloading steel coils like cylindrical rolled steel plates by inserting fastening handles into designated slots during shipping operations. The task has traditionally required teams of three workers and is considered one of the more accident-prone processes in port logistics.

HD HHI is also developing a “humanoid welding robot” with domestic robotics manufacturer Airobot, aiming to move beyond conventional collaborative welding systems.

Its strategy is to deploy humanoid robots in confined spaces inaccessible to human workers, as well as at elevated outdoor worksites where the risk of falls remains high — an effort designed to improve both workplace safety and operational efficiency simultaneously.

But experts also agree that humanoids are far from a perfect solution. The limited physical strength of current humanoid bots restricts deployment in industries such as shipbuilding and steelmaking, which require heavy loads to be lifted.

Even the very latest humanoids struggle to match the lifting capacity of an average adult man of 70 to 90 kilograms (154 to 198 pounds). Boston Dynamics’s Atlas can lift up to 50 kilograms, while Tesla’s Optimus is capable of lifting only around 20 kilograms.

Generating greater force requires larger electric motors in the robot’s appendages. But stronger motors demand thicker limbs to support the added weight, as well as larger batteries to supply sufficient power.

The result is what engineers often describe as the weight paradox — a cycle in which efforts to improve performance make the robot itself bulkier and less efficient. Ultimately, the expansion of humanoids into heavy industry may hinge on how successfully manufacturers can overcome those physical limitations.

“Rather than a single humanoid robot taking over all industrial tasks, we are more likely to see robots becoming increasingly specialized depending on the complexity of the process and the working environment,” said Prof. Cho Kyu-jin, who teaches mechanical engineering at Seoul National University.

“The key for companies will be designing robotic structures efficiently based on real-world operational data gathered from worksites.”

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.

BY KIM HYO-SEONG,JANG JOO-YOUNG [[email protected]]

Hyundai’s factory humanoid will differ from the Atlas that you know

The tumbling Atlas — developed by Boston Dynamics, Hyundai Motor’s U.S. subsidiary — is not intended for actual manufacturing use. The model that Hyundai plans to deploy in factories is the mass-produced version: the mock-up displayed at CES 2026.