NC AI and Hanwha Ocean Go After the Hardest Job in the Yard — Autonomous Welding

NC AI, the artificial-intelligence subsidiary of Korean gaming group NCSoft, has won a contract from Hanwha Ocean to develop an AI-powered autonomous welding system for shipyards — pairing computer vision with collaborative-robot control to automate one of shipbuilding's most labour-intensive and skill-dependent processes. The project, titled the development of a vision-recognition welding model and a collaborative-robot autonomous welding model, aims to move beyond conventional automation that simply repeats a pre-programmed path. What they are building: a system in which robots independently identify the weld area, assess the weld-line geometry and generate real-time control commands — driven by NC AI's industry-specialised vision-language model, expanded toward a vision-language-action controller so that a worker could instruct it in plain language rather than code. Why it is genuinely hard: shipyard welding is brutal for vision AI — intense arc light, sparks, fumes, outdoor conditions and contaminated camera lenses all degrade perception, which is why the companies are training on real worksite data and Hanwha Ocean engineer feedback. Why this matters for ship managers: you do not weld ships, but you live with the consequences — delivery schedules, build quality and the labour shortages squeezing the yards that build and repair your fleet. The move also sits inside a national push: Korea has signalled around a trillion won of public-private investment toward AI-enabled, round-the-clock shipyards targeting steep productivity gains. The honest caveat: this is an R&D-and-demonstration project, not field deployment — industry observers note the gap between lab conditions and live yards is exactly where shipyard AI tends to stall. Watch for hard metrics — weld-defect rates, cycle-time gains, system uptime in live production — before treating it as proven. [The Korea Times · Seoul Economic Daily]

This story is part of the Maritime AI Digest of 21 June 2026.