Changyeon Kim
I am a Ph.D. student at KAIST, advised by Kimin Lee and Jinwoo Shin. During my Ph.D., I was a visiting scholar at UT Austin working with Yuke Zhu and interned at NVIDIA Seattle Robotics Lab working with Yijie Guo and Yashraj Narang.
My research focuses on efficient and effective post-training for robot foundation models. I develop scalable learning, reliable evaluation, and continual adaptation methods for robots that improve through real-world experience.
Previously, I worked at Kakao. I received my B.S. in Computer Science, with a minor in Mathematics, from KAIST.
Publications
-
ACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation Models
Manuscript, 2026
Under review at ICRA 2027
-
RACE: Calibrated Cumulative Scoring for Rollout-Free Checkpoint Selection
Manuscript, 2026
Under review at ICRA 2027
-
Pretrained Vision-Language-Action Models are Surprisingly Resistant to Forgetting in Continual Learning
International Conference on Machine Learning (ICML), 2026
Oral Presentation (168 / 23918 = 0.7%)
-
DEAS: DEtached value learning with Action Sequence for Scalable Offline RL
International Conference on Learning Representations (ICLR), 2026
(^: equal advising)
-
Subtask-Aware Visual Reward Learning from Segmented Demonstrations
International Conference on Learning Representations (ICLR), 2025
(^: equal advising)
-
Guide Your Agent with Adaptive Multimodal Rewards
Conference on Neural Information Processing Systems (NeurIPS), 2023
Previously accepted to ICML 2023 Workshop on New Frontiers in Learning, Control, and Dynamical Systems
Finalist of Qualcomm Innovation Fellowship 2024 Korea
-
Preference Transformer: Modeling Human Preferences using Transformers for RL
International Conference on Learning Representations (ICLR), 2023
(*: equal contribution)