Raeyoung Kang

Gwangju Institute of Science and Technology, Gwangju University

Papers

Total Citations

H-Index

About

Raeyoung Kang is a robotics researcher pushing the boundaries of perception and manipulation in unstructured, cluttered environments. Her work centers on enabling robots to handle unseen objects—those not encountered during training—by tackling fundamental challenges in segmentation, grasping, and stable placement. Kang’s most influential contribution is her work on amodal instance segmentation, where she developed a hierarchical occlusion modeling approach that allows robots to infer the full shape of objects even when they are partially hidden. This paper has garnered 66 citations, underscoring its impact on the field. She further advanced the state of the art with GraspSAM, an innovative extension of the Segment Anything Model that enables prompt-driven, category-agnostic grasp detection, and with a large-scale simulation framework for learning to stably place partially observed objects. Kang also introduced GraspClutter6D, a large-scale real-world dataset designed to push beyond simplistic benchmarks, providing the dense occlusion and diversity needed for robust, practical grasping. Her work is essential reading for anyone interested in building robots that can perceive and act in the messy, unpredictable real world.