Doosan Robotics Selected for Two Government-Funded Physical AI R&D Projects
2026. 10. 06
- Combined R&D budget of approximately KRW 98.9 billion across two Physical AI projects
- Developing collaborative robots capable of autonomous perception, decision-making and control using domestically developed on-device AI semiconductors
- Advancing intelligent welding automation for high-complexity applications, including nuclear power equipment
Doosan Robotics is accelerating the development of collaborative robot intelligence and advanced manufacturing technologies after being selected for two government-funded R&D projects focused on Physical AI.
The company announced on Oct. 6 that its proposals for a next-generation collaborative robot and an intelligent welding solution were selected under the K-On-Device AI Semiconductor Technology Development Program and the Robot Industry Technology Development Program, respectively. The programs are jointly administered by Korea’s Ministry of Trade, Industry and Energy and the Korea Planning & Evaluation Institute of Industrial Technology.
The two projects have a combined R&D budget of approximately KRW 98.9 billion, including about KRW 68.1 billion in government funding.
The projects have distinct but complementary roles. The first focuses on foundational technologies that enhance a robot’s perception, decision-making and control capabilities, while the second applies these capabilities to high-complexity manufacturing processes. Doosan Robotics plans to integrate the technologies developed through both projects into an integrated Physical AI platform for collaborative robots.
The first project, “Development of Next-Generation Collaborative Robot Technology Based on On-Device AI Semiconductors,” aims to enable robots to perform perception, decision-making and control independently, without relying on external PCs or cloud computing.
A domestically developed AI system-on-chip (SoC) will integrate a neural processing unit (NPU) for AI computation with real-time communication and control functions and be embedded directly into collaborative robots. Using data from cameras, tactile sensors, force/torque sensors and voice inputs, the robots will be designed to make decisions and perform tasks locally in real time.
Doosan Robotics plans to develop three collaborative robot models equipped with the AI semiconductor, with payload capacities of 5 kg, 10 kg and 20 kg, as well as an industrial humanoid featuring dual-arm manipulation and omnidirectional mobility.
The systems will undergo 72-hour continuous-operation demonstrations at two or more customer sites and pursue international safety certification. Commercialization as standard products is targeted from 2031.
Once commercialized, the technology is expected to support wider adoption of collaborative robots in security-sensitive industries such as nuclear power, defense and biotechnology, where concerns over communication latency and data security can limit the use of cloud-based automation systems.
The project will run for 54 months, with Doosan Robotics serving as the lead organization. Participating organizations include Aidin Robotics, Mobilint, Korea Electronics Technology Institute, Korea Institute of Industrial Technology, SAIGE, and industry-academic cooperation foundations from Korea University, Dongguk University, Sejong University and Yonsei University.
The second project, “Development of an Intelligent Welding Solution Based on Collaborative Robots and AI,” focuses on automating high-complexity welding processes, including applications for nuclear power equipment, by combining collaborative robots with AI and digital twin technologies.
The robot will use vision systems to recognize welding targets and automatically adjust its position. Through imitation learning, it will also learn from the movements of skilled welders to generate welding paths and weaving motions autonomously.
Doosan Robotics aims to achieve more than 95% similarity between robot-generated welding paths and those of skilled welders, while maintaining a path error within ±3 mm, based on evaluation by a nationally accredited testing institution.
The company also plans to automate multi-pass welding involving five or more layers, with the goal of reducing processing time by more than 50%.
Demonstration testing will be conducted at the Doosan Enerbility Nuclear Power Center. The solution will initially be applied to processes including tack welding, preheating and welding in nuclear power equipment manufacturing. Following relevant nuclear quality certification, Doosan Robotics plans to expand the technology into shipbuilding, plant engineering and defense applications.
The project will run for 45 months, with Doosan Robotics serving as the lead organization. DeepX, SAIGE and the Industry-Academic Cooperation Foundation of Changwon National University will participate as joint research organizations, while Doosan Enerbility will take part as the end-user company.
The two projects represent an effort to integrate AI semiconductors, software and robot hardware through domestically developed technologies, extending beyond improvements in individual robot performance.
Inwon Park, President of Doosan Robotics, said, “Labor shortages and the aging of skilled workers in manufacturing are structural challenges that cannot be addressed simply by expanding conventional automation. By combining on-device AI semiconductors with our collaborative robot platform, we aim to develop Physical AI collaborative robots capable of learning and reproducing even the tacit knowledge of skilled workers, and contribute to strengthening the competitiveness of Korea’s manufacturing industry.”
