Doosan Robotics Pushes ‘Korean Physical AI’ with Homegrown AI Chips

Doosan Robotics is ramping up its development of intelligent collaborative robots equipped with domestic AI semiconductors, aiming to bring physical AI to high-tech manufacturing sectors such as nuclear power, defense and biotechnology. On October 6, the company revealed that two of its proposals have been chosen for government-funded research projects led by the Ministry of Trade, Industry and Energy and the Korea Planning & Evaluation Institute of Industrial Technology.

The combined R&D budget for these projects is around 98.9 billion won ($70 million), with roughly 68.1 billion won from the government.

The first project entails creating next-generation collaborative robots that can perceive their environment, make decisions and control their actions without relying on external PCs or cloud computing. This will involve integrating an AI neural processing unit and real-time communication and control functions into a Korean-developed AI system-on-chip for use in collaborative robots.

The robots will be built using visual, tactile, force and torque, and voice inputs to process and react to data directly at industrial sites. Three collaborative robot models with payload capacities of 5 kilograms, 10 kilograms, and 20 kilograms will be developed, along with an industrial humanoid robot capable of dual-arm operation and omnidirectional movement.

The robots will undergo rigorous testing at multiple customer locations and seek international safety certifications. Commercialization is expected by 2031.

The second project focuses on developing an intelligent welding solution that integrates collaborative robots, AI, and digital twin technology. The system will utilize cameras to recognize welding targets and allow the robot to adapt its working position based on the movements of skilled welders. AI will learn from expert welders to generate precise welding paths, ensuring minimal deviation from the skilled worker's output.

The goal is to achieve over 95% alignment between the robot-generated welding paths and those of experienced welders, with path deviation limited to within ±3 millimeters during testing by an accredited national institution. Doosan Robotics aims to automate multi-layer welding of five or more layers and significantly reduce working time by more than 50%.

The project will be conducted at Doosan Enerbility's Nuclear Power Center and will initially focus on processes such as tack welding, preheating, and welding for nuclear power equipment. Upon securing necessary nuclear quality certifications, the technology will be expanded to shipbuilding, industrial plants, and defense manufacturing.

The project will be led by Doosan Robotics, with support from DEEPX, SAIGE, and Changwon National University's industry-academic cooperation foundation, while Doosan Enerbility will act as the end-user company. Together, these projects aim to integrate AI semiconductors, software, and robotic hardware into a physical AI platform built on domestic technologies, moving beyond predefined tasks to enable robots to perceive changes in their surroundings, make decisions, and replicate complex skills typically reliant on experienced workers.

Doosan Robotics' CEO, Park In-won, highlighted the challenges posed by labor shortages and aging skilled workers in manufacturing and emphasized the potential of these technologies to boost competitiveness in Korean manufacturing.