• 강사 : 윤성의 교수 (KAIST 전산학부)
• 일시 : 2026. 4. 27. (월) 16:00~17:30

Recent advances in AI are transforming visual understanding into embodied intelligence. In this talk, I will present our recent efforts toward integrating perception, reasoning, and action through agentic AI and physical AI frameworks in robotics.
Our work builds on foundations in computer graphics and vision, extending toward multimodal understanding and decision-making systems that operate in the physical world. In particular, I will discuss how physically grounded representations (e.g., rendering-based scene understanding), multimodal retrieval, and learned world models can be combined to enable planning, interaction, and long-horizon reasoning in real-world environments.
Finally, I will highlight emerging directions in agentic systems, including memory, self-improvement, and safety, and discuss how these components contribute to building robust and adaptive embodied AI systems.