KT has begun running autonomous welding and painting robots on production work at HD Hyundai Samho in Yeongam, pairing a 5G standalone (SA) network with GPU-powered AI to control the machines in real time, according to BigGo Finance. The goal: move semi-automatic welding and painting to fully autonomous operations, raising throughput and reducing injury risk in one of the world’s most demanding manufacturing environments.
The pilot leans on three network ideas that rarely coexist on a live shop floor: SA 5G, network slicing, and AI in the radio access network (AI‑RAN). Together they give the robots a dedicated slice with priority traffic, edge inference close to the radio, and software control that can adjust to changing loads. BigGo Finance reports that KT’s setup fuses camera feeds and sensor data so the system can perceive, decide, and act with split‑second timing.
What KT AI welding robots are doing at Samho
Shipbuilding is full of repetitive motions that hide messy variation. Plates aren’t perfectly aligned. Lighting changes hour to hour. Edges warp with heat. In that noise, the KT AI welding robots track seams, adjust torch angles, and modulate current as they go. For painting, they tune spray patterns to hit coverage targets while cutting overspray. The control loop relies on GPU inference to parse video and sensor streams at the edge, then pushes commands to the actuators over a reserved 5G slice.
This is a different class of automation than classic fixture-based cells. The robots must see and adapt. That’s why the connectivity matters as much as the mechatronics. Vision models flag bead deviation, spatter, or incomplete fusion; control software closes the loop in near real time. The practical benefits, if sustained, are clear: fewer rework hours and steadier quality. Comparable welding use cases have long ranked near the top of industrial robot deployments, as tracked by the International Federation of Robotics, but shipyards have lagged fixed-line factories due to size and variability.
How 5G SA and AI‑RAN make shipyard robots viable
Wi‑Fi can cover a bay. A shipyard is a city. HD Hyundai Samho moves modules, cranes, and teams across huge outdoor spaces, which makes handoffs and interference the default state. SA 5G changes the starting point: no LTE anchor, a core designed for low latency, and slicing to give robots priority lanes and predictable quality of service. That’s useful when a weld head needs commands within tight timing windows during an arc-on cycle.
AI‑RAN, pushed by groups like the O‑RAN Alliance, moves parts of perception and control closer to the radios. Some inference can run at the edge site rather than a distant data center, so the control loop sheds travel time and backhaul jitter. For mobile robots that pass between cells and across yards, shaving milliseconds off the round trip matters as much as peak bandwidth.
There’s also isolation. With network slicing, the robots’ slice can be engineered and monitored separately from handhelds or office traffic. If a crane operator streams video, the robot slice should hold its throughput and latency targets. That separation isn’t a silver bullet, but it lets operations teams watch the KPIs that affect weld quality, rather than guessing at shared-network contention.
Why this pilot matters beyond one yard
According to BigGo Finance, the demonstration is part of South Korea’s High‑Performance AI Network Infrastructure initiative, which seeks to validate “physical AI” across shipbuilding, petrochemical, and automotive sites before wider rollout. The same report notes parallel patrol and transport trials at other industrial facilities under a separate telecom consortium. The signal is national: connect AI models to moving machines on live production, not just in labs.
The consequences reach far past a single shipyard. If weld and paint automation reaches steady-state here, expect the same recipe to target blasting, inspection, and material handling. The technology stack is modular: SA 5G for mobility and coverage, slices for predictable traffic, edge GPUs for perception, and AI‑RAN for smart scheduling and adaptation. With those pieces in place, integrating new tools becomes a software problem, which scales faster than mechanical retrofits.
What should operators and engineers watch? Three numbers stand out: defect and rework rates by weld type; arc‑on time per shift; and recordable injuries in the affected bays. If those move the right way without new downtime penalties, the business case writes itself. Training time for new hull variants will also matter. The faster teams can bring a model up to quality on a new design, the stronger the return.
Risks, costs, and what to watch next
Shop‑floor reality will test the stack. Welding throws sparks, glare, and fumes into the field of view. Fresh paint changes reflectivity. Steel blocks occlude sensors just when the robot approaches a joint. That means models must handle dynamic noise and still deliver stable seam tracking. Hardware has to shrug off heat, dust, and vibration. And networks must keep latency within tight bands when cranes or vehicles block a line of sight.
Cybersecurity isn’t abstract here. A misrouted packet can become a bad bead. Slice isolation helps, but operators will want end‑to‑end monitoring across the radio, the slice, the edge node, and the application. That’s where AI‑RAN’s telemetry could pay off: if the system can spot early drift or congestion, it can shift workloads or alert crews before quality slips.
Cost is the other brake. Edge GPUs, hardened sensors, and 5G infrastructure are not cheap. Yet shipyards face labor shortages and rising safety expectations. If the KT AI welding robots deliver lower rework hours and steadier takt without adding downtime, the spend turns into saved schedule days on a hull. For shipbuilders, days are margin.
Finally, scale depends on process fit. Painting cabins differs from painting hulls. Fillet welds behave differently than butt welds. The faster KT and HD Hyundai Samho can template those differences into repeatable software pipelines, the faster they can expand the scope without a new integration project every time.
BigGo Finance’s report lays out the core plan: prove the approach on live work, then extend it across industries under the ministry’s program. That sets a high bar, but the potential upside is real. If the KT AI welding robots can keep quality steady and uptime high under shipyard conditions, they won’t stay confined to one quay for long. For more on this, see reuters.com and bloomberg.com and nytimes.com.
Related reading: Copilot • OpenAI • Productivity & AI