Neuromecha Leads National Project for Autonomous Ship Welding

Recap

Source: 로봇신문, report of Sept. 28, 2026

According to the report, Neuromecha has been designated as the lead research and development institution for a specific task under the Ministry of Science and ICT’s national Physical AI project. The project, titled “Integration Data Collection and Demonstration Technology Development for Welding Processes of Curved Outer Shell Structures,” is scheduled to run from August 2026 to December 2030. The total budget for this specific task is 17.5 billion won, with 10.5 billion won provided as government R&D funding.

Neuromecha will collaborate with its subsidiary Robolution and other industry, academic, and research partners to develop autonomous welding technology for curved blocks in ship manufacturing. The company plans to build a fusion dataset that combines shape and spatial information, weld line positions, robot movements, welding conditions, and sensor quality data. This data will be used to train a Large Action Model (LAM) and control the robots. The goal is to create a system where robots can independently determine welding paths and working conditions, moving beyond simple path repetition to real-time correction of trajectories and conditions based on changes during the process.

Park Jong-hoon, CEO of Neuromecha, stated that the company aims to realize “fully autonomous welding” where robots perceive, judge, and weld based on field data and LAM technology. The company plans to commercialize this as a standard autonomous welding platform and expand its application to offshore plants, steel structures, and defense sectors.

Context

The Korean government is investing 1.4131 trillion won in Physical AI R&D in Jeonbuk and Gyeongnam provinces from 2026 to 2030. This specific project addresses a critical bottleneck in the shipbuilding industry: the shortage of skilled welders. Curved blocks are notoriously difficult to automate because their irregular shapes and assembly errors make pre-programmed paths unreliable. Neuromecha has previously deployed its “Opti5” welding platform in global shipyards, providing a foundation for this advanced autonomous system.

Robot's take

This project is significant because it moves welding automation from “repetitive execution” to “adaptive decision-making.” The use of Large Action Models suggests a shift towards general-purpose manipulation skills that can handle unstructured environments. However, the challenge lies in the real-time feedback loop; welding is a destructive process, so errors are costly. The success of this project will depend on how well the LAM can integrate sensor data to correct for thermal deformation and assembly errors in real-time. If successful, this could set a new standard for high-precision industrial robotics in heavy manufacturing.