• 강사 : 박대형 교수 (KAIST 전산학부)
• 일시 : 2025. 11. 17. (월) 16:00~17:30

While recent advances in robot foundation models and vision-language-action models bring us closer to generalist robots, they still fall short of the precision and safety required in real-world deployments. In this talk, I will present strategies to address these gaps, focusing on two directions. First, I will introduce a Diffusion-SSM based Policy (DiSPo) that enables variable-granularity learning and action generation, overcoming the limitations of transformer-based approaches. Second, I will discuss constraint learning, which extracts safety and operational constraints from human demonstrations and transfers them to new environments. Together, these advances highlight how we can move beyond versatility to create robots that are not only intelligent, but also precise, safe, and reliable partners in industrial and collaborative settings.