"Our ultimate goal is to implement a Physical AI factory that operates for at least 72 hours without human intervention and to develop it into a new export product for South Korea," said Kim Soon-tae, a professor in the Software Engineering Department at Jeonbuk National University and head of the Physical AI Convergence Technology Promotion Team, during a visit on August 27.
Kim emphasized, "While our goal is to create effective Physical AI software and platforms, we also need to ensure that the hardware supports this vision."
The Physical AI factory envisioned by Jeonbuk National University differs from existing unmanned factories. It aims not only to replace humans with robots but also to ensure that if one piece of equipment stops, another from a different manufacturer can take over the task and continue the process. To achieve this, the university is developing a 'Collaborative Intelligence Physical AI' platform that connects various robots and equipment into a single operating system.
During the visit to the Physical AI technology demonstration lab at Jeonbuk National University, assembly equipment surrounded by transparent safety fencing and gray robotic arms were lined up. Conveyor rails connected the equipment for moving parts, and green lights indicated operational status.
When a researcher instructed the AI factory manager to execute a process, the equipment began to operate sequentially. Autonomous mobile robots transported parts while a vision system checked their status, and robotic arms picked up and assembled the components. The actual movements of the equipment were mirrored on a digital twin screen with a slight delay of about 0.5 seconds.
Lee Jun-woo, project manager for Physical AI at the National IT Industry Promotion Agency (NIPA), noted, "For instance, if an autonomous mobile robot (AMR) from Company A breaks down, an AMR from Company B can be deployed to share operational information with surrounding equipment and continue the interrupted transport tasks, ensuring the process keeps running."
The ability to connect and replace equipment without being tied to specific manufacturers is expected to create new opportunities in the domestic manufacturing equipment industry. Currently, the AMR transporting parts in the demonstration lab is produced by the Japanese company Omron. Lee stated, "Domestic general-purpose AMRs are not widely deployed in the field, and the localization rate is low. Our goal is to increase the localization rate of general-purpose AMRs through this project."
The project team is currently using the demonstration lab as a showroom to assist domestic equipment in entering the market. Kim explained, "We can show potential demand companies the completed factory and provide equipment suppliers with opportunities to test their products in actual manufacturing processes."
Securing leadership in the factory operating platform alongside equipment localization is also a challenge. Currently, the digital twin and simulation in the demonstration lab utilize foreign solutions such as NVIDIA's Isaac Sim and Siemens' Plant Simulation. Given that Siemens control equipment is widely used in actual manufacturing sites, it is difficult to completely eliminate existing foreign technologies.
The project team plans to establish a domestic factory operating and control system (SDF-OCS) that is compatible with existing foreign equipment and software. Lee remarked, "We cannot remove all Siemens equipment installed in existing factories, so we must ensure some level of compatibility with existing equipment." He added, "We aim to validate the performance of domestic AMRs and control software before gradually replacing components."
Efforts are also underway to diversify the NVIDIA-centered computing and simulation ecosystem. The project team plans to establish a technology validation station to verify the performance and stability of Physical AI models on domestic neural processing units (NPUs).
The research team confirmed the potential for optimizing operations through the factory operating system. They identified unnecessary movements when robots transported parts and simulated improved actions in the digital twin, which were then remotely applied to the actual equipment. Kim noted that this led to a 16.7% improvement in robot movement efficiency.
The ultimate goal of the demonstration lab is to develop the factory operating platform into an export industry. To achieve this, it is essential to accumulate validation data across different industries, equipment, and processes while verifying the platform's versatility.
Lee mentioned, "Currently, there are over 200 companies and research institutions participating in the Physical AI projects in Jeonbuk and Gyeongnam. We are in the process of formalizing agreements with them and will hold a signing ceremony soon." He added, "We aim for at least 55 field applications by 2030 as part of our commercialization and technology dissemination efforts."
There are also movements to expand the validation results obtained domestically to factories abroad. Kim stated, "DH Autoliv expressed interest in applying the domestic validation results to their factory in Mexico. If we create good references domestically, we can spread the technology to the overseas factories owned by these companies."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.