Minjae Kim

I'm an M.S. student in Computer Science at Yonsei University Mirae Campus. I am currently advised by Prof. Han Ul Yoon in the HAN Lab.

I aim to build generalist robots that can robustly help people in the real world. Because continuous real-world interaction is costly and risky, my research focuses on leveraging limited, static datasets to learn broad capabilities. I believe this is best addressed by Offline Reinforcement Learning (Offline RL), and I am focusing my research on this direction.

Email  /  Github

Research Interests

I'm interested in Offline RL, model-based RL and generalist robot.

Research

Solving the One-Hole One-Ball Patience Cube Task using Model-based Reinforcement Learning Minjae Kim, Han Ul Yoon ISIS 2025   (Best paper award) In this paper, we address the OHOB patience cube task, which is one example of challenging visuomotor control problems. Therefore, we apply MBPO, known to improve sample efficiency while maintaining high performance, to solve the OHOB patience cube.

Projects

We study tactile‑driven re‑grasping: given an initial imperfect grasp, a robot should repeatedly adjust its grasp until it reaches a desired contact configuration. We collect teleoperated demonstrations with a front camera and two tactile sensors, encode observations into compact latents, and train an offline RL policy (IQL). A simple tactile success signal (“circle” marker visibility on the tactile pad) lets the policy keep trying until it achieves the intended grasp.

This project explores passive dynamic walking mechanisms through mathematical modeling and simulation. Starting with a basic compass gait model and progressing to a more complex kneeded walker, we analyze the dynamics and stability of these walking systems. Our study demonstrates how simple mechanical principles can achieve stable bipedal locomotion without active control.

This project focuses on advancing personal robotics by developing a robot capable of autonomous navigation and robotic arm control. This project demonstrates the potential of personal robots to assist with daily tasks, utilizing technologies like SLAM and computer vision without complex frameworks like ROS. The results aim to lay the groundwork for robots that can seamlessly integrate into home environments.

This project is about the two different methods for road type identification in autonomous driving. The goal of this project is to classify the road in front of the car using the camera image. To achieve this goal, I used Naive Bayes Classifier and Bayesian Estimation.