Jeonbuk National University (JBNU) Division of Electronic Engineering AI Robotics Lab (supervisor Professor Cho Hyung-gi) is broadening the horizons of Physical AI research by presenting robot intelligence technologies that understand humans and space and move autonomously in real environments.

The research team presented three research outcomes covering everything from three-dimensional perception inside the human body to space understanding that accounts for robot-specific mobility, and autonomous navigation and manipulation in real environments, at the "Nature Conference: Autonomous Robotics" held at KAIST Daejeon Main Campus from October 13–15.

This event was the first Nature Conference dedicated solely to robotics, organized by the global publisher Nature. Leading robotics scholars and Nature editors gathered to discuss the latest achievements and future research directions in autonomous robotics and Physical AI, embodied intelligence (Embodied Intelligence), robot perception and manipulation, and human–robot interaction.

Postdoctoral researcher Choi Wi-ha presented a technique that reconstructs the three-dimensional shape and motion of internal human structures from two-dimensional dynamic X-ray images. The study combines anatomical prior information with a deformable Gaussian representation to estimate dynamic three-dimensional information inside the body without direct depth data. It suggested potential applications not only in medical and rehabilitation fields but also as a new sensing technique for robots interacting with humans to assess internal human states.

PhD candidate Hyun Sang-mi demonstrated a spatial intelligence technique that represents the traversable areas for robots with differing body structures and mobility capabilities on a single map. Noting that traversability varies by robot even on the same terrain, the method considers terrain slope, step height, and passage width together with each robot's mobility. It can be used to select appropriate robots, allocate tasks, or plan routes when different types of robots collaborate.

PhD candidate Kim Kang-geon addressed the problem of a mobile manipulator—a mobile robot combined with a robotic arm—operating an unfamiliar glass door in a real environment. By analyzing constraints from the robot's direction of travel, uncertain door structures, and real actuation limits, the study outlined directions for integrated planning techniques to perform navigation and manipulation more reliably.

The through-line of the three studies is Physical AI: AI that perceives and judges the physical world and acts directly. The research demonstrated the potential to expand robot intelligence by reading internal human states, understanding space relative to a robot's physical capabilities, and acting in unpredictable real environments.

Professor Cho Hyung-gi said, "It is meaningful that the laboratory presented three diverse robotic research outcomes at this Nature Conference," and added, "We will continue to pursue robot intelligence technologies that enable robots to understand their own physical properties, their surroundings, and even human states, and to act autonomously in real environments."

Meanwhile, this research was conducted with support from the Research and Development Special Zone Promotion (R&D), the Facility Agricultural Robot Demonstration Infrastructure Construction Project, and the InnoCORE Leading-Type Project of the Science and Technology Institute.