# Taehyeon Kim (김태현)

> Taehyeon Kim (김태현) is a KAIST AI Ph.D. and Senior Research Engineer (Embodied AI) at
> Holiday Robotics. He works on embodied AI for robots that act in the real world, asking what
> knowledge a robot needs to act and designing the benchmarks and models to check it.

- Official website: https://kthyeon.com/
- Also known as: Kim Taehyeon, 김태현, Kthyeon
- Current role: Senior Research Engineer (Embodied AI), Holiday Robotics (https://holiday-robotics.com/)
- Previously: LG AI Research (Tech Lead and Research Scientist, Superintelligence Lab), Google Research, Qualcomm AI, Dynamo AI (YC W22)
- Education: Ph.D. in AI, KAIST, advised by Prof. Se-Young Yun in the OSI Lab. B.S. in Mathematical Sciences, KAIST
- Contact: kimtaehyeon610 [at] gmail [dot] com

## Current research

At Holiday Robotics, Taehyeon Kim works on embodied AI. He gives robots the knowledge they need
to act in the real world. He is not tied to one recipe like VLA or world models. He asks what
knowledge a robot needs, then designs the benchmarks and models to check it.

Current topics: skill orchestration, vision-language-action models, world-action models,
simulation benchmarks, sim-to-real benchmarks, and evaluation.

- Knowledge for embodied agents. What a robot must know before it acts, and where that knowledge comes from.
- Benchmarks and evaluation. Tests that expose real failure modes, not a fixed task suite.
- Model design. Robot policies beyond a single paradigm.
- Long-horizon autonomy. Planning, skill composition, and recovery over long horizons.

## Research background

RL and RLHF, test-time compute, search agents, perception for autonomous driving, distillation,
on-policy learning, and reward modeling. Published at ACL, ICLR, NeurIPS, and IJCAI.

## Selected publications

- Efficiently Learning To Reason or Not to Reason: Root-token Policy Optimization for Adaptive Thinking. Taehyeon Kim, Hyunsoo Lee, Youngsoo Jang, Moontae Lee. ACL 2026, Oral.
- Instructive Decoding: Instruction-Tuned Large Language Models are Self-Refiner from Noisy Instructions. Taehyeon Kim*, Joonkee Kim*, Gihun Lee*, Se-Young Yun. ICLR 2024, Spotlight.
- Exploring and Improving Multi-token Prediction (Block Draft) in Language Modeling. Taehyeon Kim, Ananda Theertha Suresh, Kishore Papineni, Michael Riley, Sanjiv Kumar, Adrian Benton. NeurIPS 2024.
- Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Federated Object Detection. Taehyeon Kim, Eric Lin, Junu Lee, Christian Lau, Vaikkunth Mugunthan. NeurIPS 2023.
- Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillation. Taehyeon Kim*, Jaehoon Oh*, Nak Yil Kim, Sangwook Cho, Se-Young Yun. IJCAI 2021.

## Links

- Google Scholar: https://scholar.google.com/citations?user=wDEaSpwAAAAJ&hl=en
- OpenReview: https://openreview.net/profile?id=%7ETaehyeon_Kim1
- GitHub: https://github.com/Kthyeon
- LinkedIn: https://www.linkedin.com/in/taehyeon-k-6a1239207/
- X: https://x.com/kimtaehyeon610
- CV (PDF): https://kthyeon.com/Taehyeon_CV_260615.pdf