NC AI Builds Autonomous Welding System for Hanwha Ocean

According to reporting by Seoul Economic Daily, Digital Today, Chosun, and Chosunbiz, NC AI secured Hanwha Ocean's project titled "Development of Vision-Based Welding-Specific Models and Autonomous Welding Models Based on Collaborative Robots." The project scope, as described in those outlets, includes a vision-recognition welding-dedicated model and a collaborative-robot autonomous welding model intended for application to next-generation commercial and special-purpose ships.

Reporting by Digital Today and Chosun states the effort will fuse AI vision recognition with precision robot control to move beyond trajectory-following automation toward robots that "recognize and judge welding areas on their own and perform optimal welding in real time." Digital Today reports NC AI will deploy its next-generation industry-specialised vision-language model, VAETKI Vision, as the project's core engine; Chosun's English coverage refers to the model as Baeki Vision and describes expansion into a vision-language-action (VLA) model that converts visual and textual/voice instructions into robot control commands.

An NC AI official quoted by Chosun outlined the operational challenges at shipyards: "The shipyard welding process is a key task that determines ship manufacturing costs and quality, but due to the nature of the work, it has extremely harsh conditions for vision recognition AI to operate, including intense arc light and sparks, welding fumes generated in real time, and camera lens contamination caused by rough field environments that include outdoor settings." Seoul Economic Daily and Digital Today add the project will use real-time worksite data and engineer feedback from Hanwha Ocean to develop shipbuilding-specialized vision recognition capable of extracting geometric weld lines and detecting defects amid noise and contamination.

Industry-pattern observations: Combining a vision-language model with closed-loop robot control to create a VLA system aligns with ongoing research trends that integrate multimodal perception and action for physical autonomy. Companies attempting similar integrations commonly face three technical hurdles: robust perception under extreme lighting and particulate conditions, low-latency mapping from perception outputs to motion/control commands, and safe, certifiable interaction between collaborative robots and human workers. For practitioners, achieving reliable weld-line extraction and defect detection in arc-heavy environments typically requires domain-specific training data augmentation, sensor fusion (for example, combining visual cameras with active sensing), and on-robot inference optimizations to meet real-time control constraints.

Editorial analysis: Public coverage frames this collaboration as part of a broader industry push to automate skilled manufacturing tasks that remain hand-intensive and quality-critical. Observers following the shipbuilding and industrial-robotics sectors note that demonstrable gains in throughput, repeatability, and defect reduction are the usual commercial levers for adoption. For AI teams, shipbuilding projects emphasize long-tail edge cases and field data collection strategies; projects that succeed tend to pair close engineering collaboration with the industrial partner and incremental validation on progressively complex assemblies.

Observers should monitor field trial results and published performance metrics such as weld-quality defect rates, cycle-time improvements, and system uptime during live production runs. Also relevant will be the sensor and model architecture choices used to harden perception against arc light and contamination, the latency between perception and actuator command, and any safety or certification steps required for deploying collaborative robots in shipyards. Finally, note the model-name discrepancy across outlets-Digital Today identifies VAETKI Vision, while Chosun refers to Baeki Vision-which could reflect translation or naming differences; further direct documentation from NC AI would clarify the product branding.