I received my master’s degree in Computer Science (MSc) from Columbia University, where I conducted research with Professor Hod Lipson and the Creative Machines Laboratory. My research focused on genetic evolutionary algorithms for processes in artificial metabolism, combining different fields in robotics, artificial intelligence, and simulation.

Prior to Columbia, I obtained my bachelor's degree in Computer Science and Engineering (BEng) from Korea University, where I worked at the Virtual and Augmented Reality Research Institute led by Professor JungHyun Han. Here, I engineered AR/VR and computer vision algorithms for efficient and compact representation of 3D geometry using dense point clouds created through real-time mobile SLAM (simultaneous localization and mapping) algorithms.

Robot Metabolism: Towards machines that can grow by consuming other machines

First robotic system capable of "robot metabolism" - physically growing from individual 1D modules into complex 3D robots by consuming and integrating material from their environment and other machines, mimicking biological organisms' ability to develop and self-repair throughout their lifetime.

Robot Links: Towards Self-Assembling Truss Robots

This work introduces Robot Links, the first truss-style modular robot capable of autonomous self-assembly and self-repair, demonstrating the transformation of independent 1D modules into fully actuated 3D robots while showcasing self-healing capabilities that reform broken connections after impact damage.

Virtual Air Conditioner's Airflow Simulation and Visualization in AR

Novel mobile AR system showcasing the first fully-interactive platform that integrates real-time 3D mapping with computational fluid dynamics simulation, enabling users to interactively install virtual air conditioners and visualize live airflow patterns and temperature changes in reconstructed indoor environments.

Python, OpenCV, Scikit-learn, Pandas, Numpy

Successfully detected car lanes from an RGB dash camera input using Canny edge detection and Hough transform in a bounded region of interest. Lane line points were fitted using an intensity histogram with a sliding window algorithm and projected onto the camera feed using perspective projection.

Python, Open3D, Numba, Skimage, Numpy

3D reconstruction from RGB-D synthetic data of the popular Disney character Wall-E. Multiple point clouds were transformed using various pose data, and TSDF (truncated signed distance function) was applied to create a 3D mesh of the character.

Python, PyBullet, PyTorch, OpenCV, Numpy

Implemented SOTA algorithm for visual robotic pack-and-place tasks. A visual affordance neural network model was trained by modifying a simplified UNet architecture to help a UR5 robot arm align the pose of its gripper. In addition, RRT (randomly-exploring random tree) was used to find a valid path from one bin to another.

C++, Eigen

Probabilistic models (Kalman Filter and Particle Filter) engineered using synthetic LiDAR and RADAR data to localize a vehicle in 2D simulation. This project was part of the Udemy Self-driving Car course.

Python, PyTorch, Scikit-learn, Numpy

Modified the neural network found in “End-to-End Learning for Self-Driving Cars,” a paper published by Nvidia for cloning the driving behavior used in autonomous vehicle research. The network was trained on augmented synthetic data obtained from a simulator.

Ruby, SQLite, Heroku

Collaborated with students from Columbia Univesity to create a full-stack SaaS (software as a service) MVP that recommended media content (movies, TV shows, games, etc) based on the people they follow on social media.