Sunpyo Hong
Papers
2
Total Citations
28
H-Index
2
About
Sunpyo Hong is a rising researcher in robot manipulation and computer vision, whose work addresses fundamental challenges in enabling robots to perceive and interact with the physical world. His primary research areas include object pose estimation, imitation learning, and robotic grasping, with a particular focus on difficult real-world scenarios. Hong’s most impactful contribution is “GhostPose” (2021, 25 citations), which tackles the notoriously hard problem of pose estimation for transparent objects—a critical gap in robot hand grasping, as standard depth sensors fail on glass and plastic. By developing a multi-view approach, he provided a practical solution for industrial and service robotics. More recently, Hong introduced the “Hierarchical Action Chunking Transformer” (2024), a novel architecture for behavioral cloning that learns temporal multimodality from diverse human demonstrations, enabling robots to replicate varied motion speeds and styles. This work directly addresses the challenge of using multi-user datasets, where different proficiency levels create complex trajectory patterns. Through these contributions, Hong is advancing the frontier of dexterous, perception-driven robot manipulation.
Research Focus
Key Achievements
Top Papers
- 1
- 2