Physical artificial intelligence (physical AI) technologies that understand the interaction between light and matter, perceive space, predict future situations, and act accordingly have been developed. They are expected to serve as a reference for implementing next-generation autonomous systems that operate in the real world, such as self-driving vehicles and humanoid robots. KAIST announced on the 6th that Professor Yoon Sung-ui’s research team in the School of Computing has developed four technologies: one that recognizes transparent objects such as glass and water, one that analyzes the interaction of light and matter to understand surrounding environments, one that enables robots to navigate to a destination using a single photograph, and one that predicts future situations to plan actions. These achievements were reported in four papers. Two were presented as oral talks and two as highlight papers at the International Conference on Learning Representations (ICLR 2026) and the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026). The team developed a visual technology for recognizing transparent environments called ‘GLINT’, enabling AI to accurately perceive transparent objects such as glass. Conventional AI systems struggle to properly separate objects reflected in glass from the scenery beyond it. GLINT separates and analyzes both the reflections on the glass and the objects behind the glass. They also developed ‘RadioGS’, a technology that understands light and material properties and reconstructs scenes. By enabling AI to grasp how light hits an object, reflects, and scatters, the system can accurately infer the material of objects and the surrounding environment even when the lighting conditions change. The team also created ‘Visual-RRT’, an image-based technology for planning robot paths, thereby finding a way to connect visual information to actual behavior. Conventional robots required coordinate data for the destination, but with Visual-RRT, the robot compares the scene it sees with a target