profile photo

Sinjae Cason Kang

  Hi 👋, I'm an undergrad @ KAIST EE, exploring Physical AI to develop self-improving robotic systems, grateful to be advised by Prof. Kimin Lee. I also actively collaborate with Kaixin Wang and Li Zhao at Microsoft Research.

Email  /  Google Scholar  /  LinkedIn  /  X (Twitter)  /  Github

Education
Korea Advanced Institute of Science and Technology (KAIST) 2020 ~ Present
B.S. in Electrical Engineering (Major), Computer Science (Minor)
  • GPA: 4.06 / 4.3 (Major: 4.12 / 4.3)
  • Served as President of the KAIST EE Student Council (2022)
  • Including mandatory military service (2023 ~ 2025)

Publications (* equal contribution)
WarmPrior WarmPrior: Straightening Flow-Matching Policies with Temporal Priors
Sinjae Kang, Chanyoung Kim, Kaixin Wang, Li Zhao, Kimin Lee
SPIGMw @ ICML 2026
project page / arXiv

Replacing the standard Gaussian source of flow-matching policies with a temporally grounded prior built from recent action history yields straighter probability paths, consistently improving manipulation success rates.

Fatigue Injection Fatigue Injection: Exposing Hidden Physical Costs via Data Poisoning in Data-Driven Robot Learning
Jian Kim*, Sinjae Kang*, JaeHyeok Doo, Kimin Lee
Under review / ArXiv Preprint, 2026
project page / arXiv

A data-poisoning attack that keeps task success intact while covertly driving robots toward far higher mechanical fatigue and energy costs, exposing a hidden physical blind spot in data-driven robot learning.

EDITH EDITH: Hierarchical Policies from Verbal and Egocentric Human Signals for Natural Human-Robot Interaction
Dongjun Lee*, Juheon Choi*, Dong Kyu Shin*, Sinjae Kang, Kimin Lee
Under review / ArXiv Preprint, 2026
project page / arXiv

A hierarchical policy framework that learns natural human-robot interaction from verbal and egocentric human signals, enabling robots to interpret spoken instructions and first-person visual cues for grounded, intuitive collaboration.

Beyond Monotonic Progress Beyond Monotonic Progress: Retry-Supervised Value Learning for Robot Imitation
Xinyao Qin, Junjie Lu, Kaixin Wang, Chuheng Zhang, Sinjae Kang, Kimin Lee, Min Xu, Bin Liang, Jun Yang, Li Zhao
Under review / ArXiv Preprint, 2026
project page / arXiv

A value-learning approach that treats retry events in imperfect demonstrations as supervision, combining global progress calibration with local preference learning to reweight demonstration segments and improve imitation from noisy human data.

Quality over Quantity: Demonstration Curation via Influence Functions for Data-Centric Robot Learning
Haeone Lee, Taywon Min, Junsu Kim, Sinjae Kang, Fangchen Liu, Lerrel Pinto, Kimin Lee
ICRA (International Conference on Robotics and Automation), 2026
project page / arXiv

A systematic method for selecting high-quality robot demonstrations using influence functions, enhanced by trajectory-level aggregation and maximum validation influence to robustly identify informative state–action data.

Automated Skill Discovery for Language Agents through Exploration and Iterative Feedback
Yongjin Yang*, Sinjae Kang*, Juyong Lee, Dongjun Lee, Se-Young Yun^, Kimin Lee^
ArXiv Preprint, 2025
project page / arXiv

An automatic skill discovery framework for LLM-powered agents that generates feasible, environment-grounded training data via exploration-first, closed-loop interaction, enabling self-improving agent capabilities without human supervision.

Research Experiences
Research Intern 2025.03 ~ Present
  • KAIST AI · RISE Lab (Advisor: Kimin Lee)
Research Intern 2022.12 ~ 2023.03
  • KAIST AI · ALIN Lab (Advisor: Jinwoo Shin)

Honors & Awards
Presidential Science Scholarship (CS field) 2020 ~ Present
  • Korea's highest honor scholarship in the sciences, granted by the President
KFAS(Korea Foundation for Advanced Studies) Undergraduate Scholarship (人材林) 2023 ~ Present
KAIST Alumni Academic Scholarship 2022 ~ Present
  • Received from Hyunggyu Lim, former President of Samsung Electronics
KAIST Insung Award 2023
  • Recognized for excellence in academics and leadership

Template