Hi, I am Junhyun Kim, an M.S. student in Robotics at Georgia Tech, advised by Prof. Zsolt Kira. My current research focuses on robot manipulation and vision-language-action models, particularly on enabling robots to learn efficiently from human demonstrations and generalize across tasks and embodiments. Previously, I worked on 3D perception, surgical robotics, and medical image restoration.
M.S. in Robotics, 2027
Georgia Institute of Technology
B.S. in Electrical and Computer Engineering, 2025
Seoul National University, Summa Cum Laude
A demo-conditioned VLA framework that improves spatial grounding through visibility-aware prediction of future end-effector traces, evaluated on RoboCasa-DC and a real-world benchmark.
Improved Activation-aware Weight Quantization with extra scaling for low-bit large language models. Obtained lower perplexity for INT3-quantized OPT and Llama 2 models.
