Taehyeon Kim (김태현)
Senior Research Engineer (Embodied AI) @ Holiday Robotics
Link to Holiday RoboticsLinks
Education
PhD — KAIST AI
BS — KAIST Mathematical Science
Minor: Intellectual Property
Bio
I am Taehyeon Kim (김태현), a Senior Research Engineer (Embodied AI) at Holiday Robotics. Previously I was a Tech Lead and Research Scientist at LG AI Research in the Superintelligence Lab. I received my Ph.D. in AI from KAIST (카이스트), where I was advised by Prof. Se-Young Yun in the OSI Lab.
I'm building embodied AI that helps robots perceive, reason, plan, and act in the real world. 로봇이 현실 세계에서 인지하고 추론하며 계획하고 행동할 수 있도록 임바디드 AI를 만들고 있습니다.
I love coffee, running, and long walks at Seoul Forest, and I enjoy the Seongsu neighborhood culture. 커피를 좋아하고, 런닝과 서울숲 산책을 즐기며 성수동 문화를 좋아합니다.
Contact: kimtaehyeon610 [at] gmail [dot] com
Official website. CV and LinkedIn are up to date.
Current Research
At Holiday Robotics, I work on embodied AI. I give robots the knowledge they need to act in the real world. 홀리데이 로보틱스에서 임바디드 AI를 연구합니다. 로봇이 현실 세계에서 행동하는 데 필요한 지식을 다룹니다.
I'm not tied to one recipe like VLA or world models. I ask what knowledge a robot needs, then design the benchmarks and models to check it. VLA나 world model 같은 특정 방법론에 얽매이지 않습니다. 로봇에게 어떤 지식이 필요한지를 묻고, 그걸 확인할 벤치마크와 모델을 설계합니다.
- 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.
Quick Answers
- Q. What does Taehyeon Kim work on in embodied AI?
- A. Skill orchestration, vision-language-action models, world-action models, simulation and sim-to-real benchmarks, and evaluation.
- Q. What is his research background?
- A. RL and RLHF, test-time compute, search agents, perception for autonomous driving, distillation, on-policy learning, and reward modeling.
- Q. How can he be contacted?
- A. Email at kimtaehyeon610 [at] gmail [dot] com. His official website is kthyeon.com.
Selected Publications
-
Efficiently Learning To Reason or Not to Reason: Root-token Policy Optimization for Adaptive Thinking
-
Instructive Decoding: Instruction-Tuned Large Language Models are Self-Refiner from Noisy Instructions
-
Exploring & Improving Multi-token Prediction (Block Draft) in Language Modeling
-
Navigating Data Heterogeneity in Federated Learning A Semi-Supervised Federated Object Detection
-
Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillation
† co-corresponding authors; * equal contribution.
Honors
- Technical Reviewer, ICML 2026
- Google Conference Scholarship For NeurIPS 2024 multi-token prediction work.
- Winner, Qualcomm AI Fellowship FINE Samples for Learning with Noisy Labels (NeurIPS '21).
- NeurIPS competition awards Weather4Cast 4th; Black-Box Optimization 8th; MicroNet CIFAR-100 2nd & 3rd.
Leadership
- Tech Lead & Research Manager, LG AI Research Led research across retrieval, LLM alignment, and agentic systems.
- Doctoral Student Representative, KAIST AI
- Lab Master, OSI Lab
- Vice Captain, KAIST ELKA