We introduce a method to optimize Gaussian splatting with a limited image set, mitigating overfitting by incorporating a estimated depth map as a geometric guide.
We study the incremental learning scenario in Neural Radiance Fields (NeRF) first time and propose a self-distillation-based method alleviating catastrophic forgetting without memory increasing.
We propose a precise navigation system using wheel odometry, Low-Grade inertial measurement unit, and monocular camera in Urban Area. VIO Simluation code (Pended)
We develop whole solid sounding rockets from engine, fuel, body, electric system including wing control, parachute control, ignition control, data(pressure, accel, rotation) measurement, and communication system.
I was a leader of electric system team. (2016 ~ 2017)
Alpha-D: Developing drone controlled by reinforcement learning
Yongjae Kim, Jaeyoung Chung, Joonyoung Jang, Chanyoung Jung
Spring ~ fall 2016
code (ROS)
/
code (GPS/INS, Kalman Filter)
We develop the whole system of quadcopter drone from scratch and try to control some complex movement by utilizing reinforcement learning.
We develop an electronic bicycle with an interchangeable front wheel between motor-powered wheel and normal. We design the motor control system deliberately for similar user experiences with pedal assistance system (PAS).
I mainly develop embedded system and algorithm including power, circuit, PID control, etc.
X-mas Tree for Nerds Jaeyoung Chung
Chrismas season, 2015
I made this tree out of discarded materials including green pcb-boards (as leaf), shiny light LED arrays, a drone that reacts when approached, and a blower that floats a ping-pong ball