Dohyun Lee

Ph.D. Student, KAIST AI

DAVIAN Lab KAIST AI

Hi! I am a Ph.D. student at KAIST AI (DAVIAN Lab), advised by Jaegul Choo.

My research focuses on robot learning, in three directions: (i) VLA and WAM Training (FUMA, PointmapVLA): learning objectives and representations; (ii) Reinforcement Learning (SVFM): RL for robots, including sim-to-real transfer and human-in-the-loop learning; and (iii) Agentic Robotics: directly predicting end-effector actions based on strong reasoning.

Dohyun Lee

News

Publications

/

Future Matching: Improving Your Vision-Language-Action Models via Differentiable Latent World Model
Dohyun Lee, Kinam Kim, Jaegul Choo, and Minho Park
Preprint 2026.
paper / project page
Animation: SVFM rolls out behavior continuations from an anchor, scores their endpoints with the critic, combines them into a soft value, and moves the actor along its gradient.
Soft-Value Flow Matching for Offline Reinforcement Learning
Dohyun Lee*, Junhyeong Hong*, Theo Choi, Kinam Kim, Jaegul Choo, Kyungjae Lee†, and Yoonseon Oh†
Preprint 2026.
paper / project page
PointmapVLA: Scaling Robot Learning with Robot-Centric Pointmaps
Dongyoon Hwang*, Dongjin Kim*, Byungkun Lee*, Kyungmin Lee, Dohyun Lee, Hojoon Lee, Minho Park, and Jaegul Choo
Preprint 2026.
paper / project page / code
An Agentic Framework for Proactive Robots with Procedural Assistance Memory
Dohyun Lee*, Minho Park*, Jeonghoon Park, Dongyoon Hwang, Junha Hyung, Sungwon Hwang, Byungkun Lee, and Jaegul Choo
CoRL 2026.
paper / project page / code
PersonaTrail: Benchmarking Personalized Web Agents through Browsing Trails
Seungbin Yang, Chaewoon Ki, Dohyun Lee, Jaegul Choo, and ChaeHun Park
arXiv 2026.
paper
Delving into Large Language Models for Effective Time-Series Anomaly Detection
Junwoo Park, Kyudan Jung, Dohyun Lee, Hyuck Lee, Daehoon Gwak, ChaeHun Park, Jaegul Choo, and Jaewoong Cho
NeurIPS 2025.
paper
Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transport
Minseok Choi, Daniel Rim, Dohyun Lee, and Jaegul Choo
ACL 2025.
paper
Revisiting LLMs as Zero-Shot Time Series Forecasters: Small Noise Can Break Large Models
Junwoo Park, Hyuck Lee, Dohyun Lee, Daehoon Gwak, and Jaegul Choo
ACL 2025.
paper
Exploring In-context Example Generation for Machine Translation
Dohyun Lee, Seungil Chad Lee, Chanwoo Yang, Yujin Baek, and Jaegul Choo
Findings of ACL 2025.
paper
Breaking Chains: Unraveling the Links in Multi-Hop Knowledge Unlearning
Minseok Choi, ChaeHun Park, Dohyun Lee, and Jaegul Choo
arXiv 2024.
paper
PairEval: Open-domain Dialogue Evaluation with Pairwise Comparison
ChaeHun Park, Minseok Choi, Dohyun Lee, and Jaegul Choo
COLM 2024.
paper
Protecting Privacy Through Approximating Optimal Parameters for Sequence Unlearning in Language Models
Dohyun Lee*, Daniel Rim*, Minseok Choi, and Jaegul Choo
Findings of ACL 2024.
paper

Work Experience

KRAFTON AI
Research Intern, AI Companion Team
  • Built an AI co-playable character (CPC) for PUBG Ally on top of LLM/VLM-based agent systems. demo
  • PUBG Ally was showcased at CES 2025 in collaboration with NVIDIA, and beta-launched in PUBG: BATTLEGROUNDS.
Hyundai Motor Company
Undergraduate NLP Researcher

Education

KAIST
Integrated M.S./Ph.D. in Artificial Intelligence (Advisor: Prof. Jaegul Choo)
Kyung Hee University
B.S. in Mechanical Engineering and Software Engineering
4 Academic Excellence Scholarship, 2 Academic Excellence Award