Superhuman level performance across all the multi-agent atari games in PettingZoo (Feb. 2022 - Jun. 2022)
Teammates - J K Terry (Maintainer of OpenAI Gym), Kevin Corder, Costa Huang, Jaekyung Cho, Eunki Jung
Role - code review & troubleshooting (SuperSuit, Stable-baseline3, PettingZoo), Checkpoint on cluster computing for OPtuna
Training a customized performance low-level controller for a quadrotor in a Sim2Real manner (Jan. 2022 - Current)
International research internship (Human Resource Development Program for Industrial Innovation)
University of Maryland, Motion and teaming lab (Dr. Machael Wilson Otte)
Fully funded by Korea Institute for Advancement of Technology(KIAT) grant funded by the Korea Government(MOTIE)
Teammates - Jaekyung Cho, Chan Kim, Mohamed Khalid, Michael Otte
Role - Project leader, Full stack (Hardware, Software) engineer
2021
Blind spot detection through sound (Jul. 2021 - Current)
Autonomous Robot Intelligence Lab (ARIL), SNU
Teammates - Jaekyung Cho, Bonhak Koo
This is the project to detect the position of occluded vehicles by walls on narrow alley roads through sound. We used beamforming method to determine probability that a sound source (generated by occluded vehicles) exists in a specific direction and applied particle filter method based on that probability.
Solving snake game with reinforcement learning (Jun. 2021 - Feb. 2022)
This is an educational program to help children exercise in a pandemic situation. Pose estimation is conducted through the PoseNet and multiple games were designed for encouraging students to exercise.
2020
Motion Planning in CARLA Simulator (Aug. 2020 - Mar. 2021)
This is the project that developed a novel hybrid motion planning method for an autonomous vehicle which combines single-query method and multi-query method, and verified in the CARLA simulator
Autonomous Golf Car (Jun. 2020 - Jul. 2021)
Autonomous Robot Intelligence Lab (ARIL), SNU
Teammates - Jaekyung Cho, Chan Kim, Donghwi Jung, Hyosang Ahn
Role - Project leader, hardware re-factoring, motion planning and control
This is the project that re-factorizing a golf car to be able to autonomously drive. The steering wheel was controlled by an additional DC motor and the gas pedal was controlled by Arduino with hijacking motor signal. Multiple sensors (16-channels LiDAR, stereo cameras, IMU, GPU, and wheel encoders) were mounted in proper position. All systems were connected using ROS in the main computer (Jetson Xavier)
Autonomous Parking with RRT in Unity3D (Jan. 2020 - Mar. 2020)
Autonomous Robot Intelligence Lab (ARIL), SNU
Personal project
Role - Unity environment design, Path planning algorithm implementation
Motion planning by RRT* algorithm and Dubins path, and control by pure pursuit algorithm. This is my first project in the autonomous vehicle field and ARIL.
2019
Human following control of holonomic robot based on vision recognition (Sep. 2019 - Aug. 2020)
Undergraduate graduation thesis, SNU MAE
Personal project
Role - Robot design & manufacture, deep learning engineer
This is the graduation thesis project for Mechanical engineering departmen, Seoul National University. I designed a holonomic mobile robot by mechanum wheels that operated by ROS. OpenCV library was used to recognize a human size and to estimate distance from the robot.
Autonomous Emergency Braking for Bicycle (Sep. 2019 - Dec. 2019)
Mechatronics, SNU MAE
Teammates - Jaekyung Cho, Seungyoon Cho, Soojin Lee, Chayeon Lee
Role - Project leader, deep learning engineer, re-factorizing hardware
This is a driving assistance systems for bicycles. It detects the person in front by Yolo-v3 and gives differential warnings according to the distance. In case of hazardous situation, the automatic brake is operated with the LED warning system.
Role - Mechanical engineer, 3D modeling, dynamics analysis
This is the project that Re-factorizing a segway to be able to climb stair. The optimal caterpillar tracks structure was found through 3D modeling and dynamics analysis.