Biography
I am currently a Robotics Software Engineer at AI & Robotics Ventures, a start-up from a petroleum company named PTTEP. My work revolves around pushing the autonomous level of the subsea pipeline inspection robot, by solving some of the challenging problems in computer vision, robotics, and deep learning domain.
I graduated from Korea Advanced Institute of Science and Technology (KAIST) with an Electrical Engineering major and a Computer Science minor, where I had several interests from computer architecture, CUDA programming, computer vision, to robotics (in chronological order). I am thankful for Vertically Integrated Architecture Research Group for allowing me to participate in the research during my early year, which helped me discovering myself that I was not a good computer architect, but rather that I was interested in writing efficient C/C++ code for hardware like GPUs. I was indebted to Urban Robotics Reserach Group for supporting me throughout the participation in the final year, where I worked on LiDAR-based 3D object detection for a self-driving car competition in South Korea.
I believe in the future of autonomy and that wakes me up every morning. I am passionate about the digital transformation and about building software infrastructure that will help create the autonomous world.
Outside of work, I enjoy playing badminton and hiking. I also participate in data science competition occasionally, but I have been super busy since joining the current start-up.
Download my resumé.
Education
Experience
Autonomous Underwater Vehicle for Subsea Pipeline Inspection
- Lead 1 AI project. Implement object detection and tracking, and geometrical path planning algorithm, which improves the auto docking process by significantly reducing the operation time.
- Optimize all deep learning models currently used in the robot for real-time inference. Keywords: model quantization, Intel OpenVINO, Nvidia TensorRT
- Work on anomaly detection using synthetic data. Currently working on dataset generation in Unity.
- Create a GUI tool that helps us tune 3D computer vision algorithms in an interactive way.
- Develop machine leaning models for automatic pipeline inspection. Write a custom 3D pointcloud segmentation/clustering algorithm which improves the robot navigation.
- Implement a custom sonar simulation in Gazebo using computer vision techniques, allowing us to verify the whole algorithm in simulation without having to go offshore. The algorithm is written in CUDA, ~2-3x faster than OpenCV.
- Implement sensor fusion based on EKF.
- Write and maintain serveral hardware drivers interfacing with the AUV, including INS and USBL communication system.
- Write and migrate software stack from ROS1 to ROS2.
- Participate in the research discussion at Kamijo Lab
- Worked with a team to develop a perception software for a real car, as part of a Self-Driving Competition in South Korea.
- Trained a deep learning model for LiDAR-based 3D object detection. Optimized for the inference by using ONNX and Nvidia TensorRT, running ~18 FPS on a laptop-grade GPU.
- Proposed the data augmentation technique, increasing the accuracy by ~10%.
- Implemented 3D tracking using Kalman filter (tracking by detection).
- Implemented a custom-designed accelerator for a sparse CNN model using Vivado High-Level Synthesis Language (HLS).
- Tried to hack into TVM (Open Deep Learning Compiler) and use it to compile a deep learning model into a hardware code that runs on a custom-designed FPGA-based accelerator.
- Implemented PID control software on Renesas RX62T controller board.
- Worked on path following simulation in MATLAB.
- Studied analog circuits for energy harvesting devices.
Skills
c++11, c++14, template
Python 2, 3
ROS1, ROS2
(Basic)
(Basic)
(Basic)
(Basic)
(Basic)
(Novice)
(Novice)
(Basic)
Tools
Certificates
Recent Posts
Gallery
