Kun-Yu Lin

Papers

Total Citations

H-Index

About

Kun-Yu Lin is a rising researcher at the intersection of computer vision, robotics, and embodied AI, whose work focuses on bridging the gap between visual perception and real-world robotic manipulation. His key research areas include task-oriented grasping, human-robot domain adaptation, and procedural error detection. Lin’s major contributions center on developing visual representations that generalize across diverse embodied environments—a critical challenge given the limited scale of robot demonstration data. His 2025 paper on mitigating human-robot domain discrepancies in visual pre-training (4 citations) proposes novel methods to leverage large-scale human video data for robotic tasks, significantly reducing the need for expensive robot-specific training. In task-oriented 6-DoF grasp pose detection (4 citations), Lin addresses the nuanced problem of context-dependent grasping, where an object’s intended use dictates how it should be held—a key advancement for assistive robotics. His work on modeling multiple normal action representations for error detection (2 citations) introduces dynamic prototypes to identify anomalies in procedural tasks, enhancing reliability in AR-assisted and autonomous systems. Though early in his career, Lin’s innovative approaches to representation learning and task-aware manipulation are already shaping the next generation of adaptive, human-aware robotic systems.