Jaeuk Shin

Hi! I am currently pursuing a Ph.D. in Electrical and Computer Engineering at Seoul National University, where I am a member of LDS lab and advised by Prof. Insoon Yang. My research focuses on the theory and practice of learning-based decision making and control. I received my B.S. and M.S. degrees (advisor: Prof. Myungjoo Kang) in Mathematics from Seoul National University.

My Research lies at the intersection of control theory, reinforcement learning, and robotics. Broadly, I study how data-driven approaches and principled optimization can be combined to enable safe, efficient, and adaptive autonomy in complex environments. My work has explored topics including model predictive control, meta-reinforcement learning, conformal prediction for safety-aware navigation, and structural analysis of multi-task MDPs. I am particularly interested in devloping methods that integrate prediction, uncertainty quantification, and control to build reliable real-world robotic systems.

Email / CV / Scholar / LinkedIn / Github

Education

Ph.D. in Electrical and Computer Engineering, Seoul National University, 2020–2026
M.S. in Mathematics, Seoul National University, 2018–2020
B.S. in Mathematics, Seoul National University, 2014–2018

Research

	From Prediction Uncertainty to Conformalized Distance Fields for Safe Motion Planning Jaeuk Shin, Yoonseok Ra, Insoon Yang preprint, 2026 Paper We develop a safe motion planning algorithm that conformalizes the entire distance field based on functional conformal prediction, a functional analytic variant of the standard conformal prediction.
	CANVAS: Competency-Aware Navigation Assessment Suite for Pedestrian Trajectory Forecasting Jaeuk Shin, Joonho Han, Gihwan Kim, Jungjin Lee, Insoon Yang IEEE Open Journal of Control Systems (OJ-CSYS), 2026 Paper Github We design control-aware evaluation metrics for pedestrian trajectory forecasting models based on conformal prediction and a simulation testbed for their assessment.
	Koopcast: Koopman Operator-Based Trajectory Forecasting Jungjin Lee, Jaeuk Shin, Gihwan Kim, Joon Ho Han, Insoon Yang IEEE Conference on Robotics & Automation (ICRA), 2026 Paper We develop a Koopman-operator-based pedestrian trajectory forecasting model that achieves fast inference while maintaining competitive performance.
	Egocentric Conformal Prediction for Safe and Efficient Navigation in Dynamic Cluttered Environments Jaeuk Shin, Jungjin Lee, Insoon Yang IEEE Conference on Decision and Control (CDC), 2025 Paper Github We develop a conformal prediction scheme for direct safe-set calibration that maintains theoretical guarantees while improving efficiency over prior CP-based control methods.
	On task-relevant loss functions in meta-reinforcement learning Jaeuk Shin, Giho Kim, Howon Lee, Joonho Han, Insoon Yang Annual Learning for Dynamics & Control Conference (L4DC), 2024 Paper Github Incorporating value-function information into the model-learning objective significantly accelerates meta-RL.
	Multiparametric Analysis of Multi-Task Markov Decision Processes: Structure, Invariance, and Reducibility Jaeuk Shin, Insoon Yang IEEE Control System Letters, 2024 (Selected for presentation at CDC'24) Paper We present a geometric perspective on understanding multi-task Markov decision processes. The resulting structure of the task space provides a unifying view of reward shaping and task identification, enabling us to address questions arising in zero-shot learning.
	Anderson acceleration for partially observable Markov decision processes: A maximum entropy approach Mingyu Park, Jaeuk Shin, Insoon Yang Automatica, 2024 Paper Github Soft regularization of approximate POMDP operators enhances the effectiveness of Anderson acceleration.
	Control of fab lifters via deep reinforcement learning: A semi-MDP approach Giho Kim, Jaeuk Shin, Gihun Kim, Joonrak Kim, Insoon Yang IEEE Transactions on Automation Science and Engineering (T-ASE), 2023 Paper Github A semi-MDP formulation of Q-learning enables the solution of large-scale industrial problems. (collaboration with SK Hynix)
	Infusing model predictive control into meta-reinforcement learning for mobile robots in dynamic environments Jaeuk Shin, Astghik Hakobyan, Mingyu Park, Yeoneung Kim, Gihun Kim, Insoon Yang IEEE Robotics and Automation Letters (RA-L), 2022 (Selected for presentation at IROS'22) Paper Github We present MPC-PEARL, a novel meta-RL scheme that leverages MPC to rapidly generate faithful yet complex local behaviors during training. This compensates for the slow initial learning of meta-RL algorithms, while its myopic tendency can be addressed through RL.
	Hamilton-Jacobi deep Q-learning for deterministic continuous-time systems with Lipschitz continuous controls Jeongho Kim, Jaeuk Shin, Insoon Yang Journal of Machine Learning Research (JMLR), 2021 Paper Github Starting from continuous-time HJB equations, we derive HJDQN, a continuous-time Q-learning method. This eliminates the need for a separate actor module and ensures smooth action trajectories.

Experience

Research Intern
NAVER LABS, Gyeonggi, South Korea
December 2025 – June 2026

Development of an RL-based visuomotor policy for end-to-end autonomous urban navigation, integrating visual perception with control in a mobile robotics platform.

Academic Services

Reviewer for Journals

IEEE Transactions on Robotics (T-RO)
IEEE Transactions on Automation Science and Engineering (T-ASE)
IEEE Robotics and Automation Letters (RA-L)
IEEE Control System Letters (L-CSS)

Reviewer for Conferences

International Conference on Learning Representations (ICLR)
ACM International Conference on Hybrid Systems: Computation and Control (HSCC)
American Control Conference (ACC)
IEEE Conference on Decision and Control (CDC)
Annual Learning for Dynamics and Control Conference (L4DC)
Conference on Robot Learning (CoRL)