DeepX, a South Korean artificial intelligence chipmaker, is developing an AI-based intelligent welding robot solution with Doosan Robotics to address a shortage of skilled welders at nuclear power plant manufacturing sites.
DeepX said on the 6th that it was selected for the Ministry of Trade, Industry and Energy's 2026 Robot Industry Technology Development Project and will begin joint development with SAGE and the Industry-Academic Cooperation Foundation of Changwon National University.
The project centers on combining vision AI with collaborative robot technology to automate welding processes. The consortium will develop technology that uses AI to recognize and correct welding targets and paths, and that analyzes vision and thermal imaging data collected during welding to assess weld quality and detect defects. The plan is to pair that capability with collaborative robots to create an intelligent welding solution that can be deployed at actual manufacturing sites.
Doosan Robotics, the lead research and development institution, will oversee the consortium and design the system integration architecture, developing a collaborative robot welding cell, AI for correcting micro-displacement and real-time welder control technology. DeepX will develop on-device edge AI hardware based on a domestically produced neural processing unit (NPU) along with a software development kit (SDK). It will also build a three-channel vision data pipeline, integrate the system with robot controllers, develop thermal imaging analysis AI models, and support environmental resistance testing and reliability certification for welding conditions. SAGE will handle 3D vision-based recognition of welding targets, welding path correction and AI models for detecting weld bead defects, while the Changwon National University foundation will develop AI-based welding path generation and physics-based digital twin technology.
The technologies developed separately by the four organizations will be integrated into a single intelligent welding system. Phased demonstrations are planned at Doosan Enerbility's nuclear power equipment manufacturing sites during the project period. The partners aim to complete quality verification for nuclear power equipment in 2029 and begin commercialization in 2030.
"This project will be a concrete case of physical AI, in which ultra-low-power AI chips are combined with robots and used at actual manufacturing sites," DeepX said. "Based on DeepX's NPU and on-device AI technology, we will contribute to building an AI computing environment that can efficiently process and reliably run the AI computations required at manufacturing sites."
