Heungwoo Han

Papers

Total Citations

H-Index

About

Heungwoo Han is a roboticist focused on advancing robot perception and manipulation, particularly in complex, real-world environments. His research centers on three core challenges: pose estimation for difficult objects, learning from human demonstrations, and robust grasping. Han’s most impactful work, "GhostPose" (2021, 25 citations), tackles the notoriously difficult problem of estimating the pose of transparent objects—a critical gap for robot hand grasping, as standard depth sensors fail on glass or plastic. This contribution directly addresses a key bottleneck in industrial and service robotics. He further advances imitation learning with his "Hierarchical Action Chunking Transformer" (2024), which enables robots to learn multimodal behaviors, like varying speeds, from diverse human demonstrations, overcoming a major hurdle in behavioral cloning. Complementing these, his "RGBD Fusion Grasp Network" (2023) introduces a large-scale dataset and a novel method for stable grasping of flat tableware in home settings. Together, Han’s work systematically pushes the boundaries of how robots perceive and interact with the physical world, from transparent objects to everyday dishes, laying essential groundwork for more capable and adaptable robotic assistants.