Learning Multimodal One-step Flow Policy via Value-weighted Optimal Transport
NeurIPS 2026 Conference Submission
arXiv
Paper
Hi, I'm Jinha Choi
I am an Undergraduate Researcher
About
Undergraduate Research Intern @ Yonsei University / Senior Student @ Yonsei University
Research Intern in AI Graduate School
I am Jinha Choi, working as a research intern at DILLAB (Yonsei Univ) advised by Prof. Jongmin Lee. My current research focuses on learning efficient multimodal flow policies via value-weighted optimal transport, distilling a value-aware multi-step reference policy into a one-step target through entropic optimal transport.
More broadly, I am interested in robot learning for complex embodied systems — particularly humanoid behavior, whole-body control, and manipulation. My goal is to develop scalable and principled methods that allow humanoid robots to learn robust, adaptive, and generalizable behaviors for challenging real-world tasks.
Research Interests
Reinforcement Learning
Action-constrained imitation/reinforcement learning, flow matching
Robot Learning
Humanoid loco-manipulation, whole-body control
Experience
-
HuGeLab (KAIST AI) Jun 2026 ~Research Intern
Advised by Prof. Beomjoon Kim -
DILLAB (Yonsei Univ.) Jan 2026 ~ Jun 2026Research Intern
Advised by Prof. Jongmin Lee -
RLLAB (Yonsei Univ.) Jul 2025 ~ Dec 2025Research Intern
Advised by Prof. Youngwoon Lee -
MIRLAB (Yonsei Univ.) Sep 2024 ~ Dec 2024Research Intern
Advised by Prof. Youngjae Yu -
Military Service Mar 2023 ~ Sep 2024Mandatory Military Service
Korea Defense Intelligence Command -
Yonsei University 2021 ~ 2026 (Expected)B.S. Student
Nano Science & Engineering | Computer Science -
West Vancouver Secondary School Sep 2015 ~ Jun 2020International Baccalaureate Diploma
WVSS IBDP | West Vancouver, Canada
Robotics
VEX Robotics — Team 1010B
1010B Team Lead · WVSS Robotics Academy · Sep 2017 – Jun 2020
Competed across VEX Robotics Competition seasons Starstruck, In the Zone, and Turning Point. Led Team 1010B (6–12 members) through the full robot-development cycle: game strategy analysis, mechanical and electrical design and build, programming motor/sensor control and autonomous routines, and competitive driving. Coordinated with alliance teams during the World Championship.
1st Place
VEX Robotics Canada BC Provincial Championship
British Columbia, Canada · 2019
Finalist
VEX Robotics World Championship
Kentucky, USA · 2019
Physical AI 26H Robotics Hackathon
Hosted by Kernel Academy & Doosan Robotics · Mar 2026
Built a natural-language-guided pick-and-place system on a Doosan E0509 robot arm using RGB-D sensing, vision-language reasoning, and ROS2 execution. Spoken commands were transcribed with Whisper, scene objects segmented with SAM2, and pick/place targets resolved through GPT-4o reasoning. End-effector coordinates were localized from depth and executed via calibrated camera-to-robot transformation and motion primitives, with before/after verification and error-recovery logic for reliability.
2nd Place
Physical AI 26H Robotics Hackathon
Kernel Academy & Doosan Robotics · 2026
Honors & Awards
2024
Top Award, Korea Defense Intelligence Command AI Security Competition
Received a top award in an AI security competition. Proposed an on-device AI framework for military systems with partial internet exposure, designed to support operational convenience while strengthening protection against security vulnerabilities and malicious misuse.
2024
Peer Mentor, Yonsei–Nexon RC Creative Platform
Selected as a peer mentor based on prior competition performance (former 1st-place team). Provided guidance to participating teams and received an official mentoring certificate.
2022
Honor Roll
Awarded in recognition of commendable academic achievement.
2021
1st Place, Yonsei–Nexon RC Creative Platform
Won 1st place in a university-wide social-impact startup ideation competition co-hosted with Nexon. Led a 5-member team to design an automated revolving-door assistance system to improve safe building access for people with mobility impairments. Developed a 3D-printed prototype and presented the design rationale (safety, practicality, community impact) to judges and organizers; ranked 1st of 90 teams and awarded $10K.
Contact
Yonsei University, Yonsei Engineering Research Park, Room 353A
Sinchon, Seoul, Republic of Korea
Sinchon, Seoul, Republic of Korea
jinhac174@gmail.com
LinkedIn