Seoul, South Korea, October 2, 2025: South Korean researchers have developed a new robotics system that mimics human memory processing to improve the performance of autonomous mobile robots in industrial environments. The technology enables robots to prioritize real-time data and discard outdated information, enhancing navigation efficiency in logistics centers and smart factories. The study was conducted by the Daegu Gyeongbuk Institute of Science and Technology (DGIST) and published in the Journal of Industrial Information Integration.
The research introduces a “Physical AI” model that draws on a concept of “spread and forgetting,” inspired by the way social issues emerge and fade over time. This approach allows autonomous robots to filter out obsolete data, such as previously encountered obstacles that no longer exist, thereby avoiding unnecessary detours and optimizing task flow. Professor Kyung-Joon Park of DGIST’s Physical AI Center led the research team, which included Jiyeong Chae and Sanghoon Lee. The team focused on improving the cooperative navigation capabilities of Autonomous Mobile Robots (AMRs), which are commonly used in manufacturing, logistics, and warehouse operations.
Conventional navigation systems often cause robots to reroute around former obstructions, reducing operational productivity even after the obstacle has been cleared. To test the new system, the researchers used the Gazebo simulator to recreate a logistics environment. The performance of the Physical AI model was benchmarked against the widely used ROS 2 (Robot Operating System 2) framework. The new model showed a reduction in average driving time by up to 30.1 percent and an increase in task throughput by up to 18 percent.
South Korea expands robotics research with new alliances
According to the researchers, the model enables robots to share only high-priority data, such as the location of a current obstruction, while gradually forgetting information that is no longer relevant. This selective data sharing is designed to reduce communication overload within robot fleets and to improve overall coordination without requiring external computation or cloud processing. The system operates using 2D LiDAR sensors, eliminating the need for expensive additional hardware.
It has been developed as a plugin for ROS 2, allowing for straightforward integration into existing robotic platforms. The researchers emphasized that the approach is designed to work in real-time, with no reliance on pre-mapped static environments. In addition to improved navigation, the system offers potential benefits in reducing energy consumption and mechanical wear by avoiding inefficient routing and unnecessary stops. These improvements could contribute to lower operational costs in high-throughput environments where delays and equipment fatigue are significant factors.
Task throughput improved with selective memory model
The technology has been designed for immediate industrial use and is compatible with current robotics infrastructure. While the initial tests were conducted in simulated environments, the researchers noted that the plugin is now available for application in commercial AMR systems. This development comes amid increased interest in robotics automation in South Korea. Earlier this year, the country launched the K-Humanoid Alliance, a national initiative aimed at coordinating research in robotics and artificial intelligence across academic, industrial, and government sectors.
The DGIST research adds to a growing portfolio of innovations aimed at making autonomous systems more adaptable and efficient in real-world operations. The Physical AI-based model represents a shift in how robots process and act on environmental data, with a focus on real-time decision-making and operational precision. The research team has made the plugin publicly accessible to facilitate adoption across a range of industries including logistics, manufacturing, and autonomous systems development. – By Content Syndication Services.