π€ Building robots that can perceive, decide, and act.
Robot Perception β Simulation & Manipulation β Action Intelligence μ°μ νκ²½μμ μ λ’°ν μ μλ Physical AIλ₯Ό μ°κ΅¬νκ³ κ΅¬νν©λλ€.
π§ Research Direction
|
ποΈ Perception 6D Pose / 3D Vision RGB-D Β· FoundationPose Part-level Segmentation Stereo Vision Β· Point Cloud |
β |
π¦Ύ Manipulation Simulation / Control CoppeliaSim Β· IK FSM Β· Grasping Sim-to-Real |
β |
π§ Action Intelligence VLA / Robot Foundation Model GR00T N1.7 Diffusion Β· Attention OOD Robustness |
π¬ Current Research
GIST REAL-Lab β Diffusion-based Vision-Language-Action Research
Goal: Improve robot-action stability and robustness by controlling the inference process of pretrained VLA models.
Analyze Vision / Language / State / Action interactions in GR00T N1.7
Explore two complementary intervention directions:
Cross-Attention β sparse retrieval of task-relevant visual / language conditions
Self-Attention β symmetric / skew decomposition for analyzing robot-action dynamics
Evaluate inference-time intervention on LIBERO+ under language, robot, and noise perturbations
Build pipelines for model modification, large-scale simulation, ablation, and success / failure analysis
π Industrial Robotics Research
|
ETRI β Robust Part-Level 6D Pose Estimation Problem Chrome-plated VCB components caused severe depth corruption due to specular reflection, making standard RGB-D pose estimation unreliable. Approach Target-specific part segmentation FoundationPose-based pose estimation Geometry-aware pose filtering Mask-based pose re-evaluation Real robotic manipulation validation Result Rotation MAE: 21.67Β° β 2.43Β° Translation MAE: 1.25 cm β 0.88 cm Verified feasibility in a real robotic manipulation environment |
π Publication Robust Part-Level 6D Pose Estimation for Specular Metal Components using Mask-Optimized FoundationPose IEEE Access Co-author, 2026 |
π¦Ύ Robotics Projects
Stereo Vision-based UR5 3D Pyramid Stacking
Stereo Camera β Color Detection β 3D Triangulation β IK β FSM β UR5 + RG2
Built an end-to-end manipulation pipeline in CoppeliaSim
Reconstructed block positions using epipolar geometry + stereo triangulation
Achieved approximately 1 cm localization error
Controlled UR5 using simIK and an FSM-based pick-and-place controller
Improved physical grasping with collision avoidance, gripper pose constraints, friction tuning, and grasp-depth tuning
GuideAI β Real-time 3D Vision for Visually Impaired Pedestrians
Designed a Two-Track perception pipeline
Structured objects β 6DoF Pose Estimation
Unstructured objects β Segmentation + Depth Estimation + Point Cloud
Integrated obstacle distance / direction estimation with voice and vibration guidance
Evaluated in real campus walking environments
Presented at the 2026 Smart Media Spring Conference
π§ͺ Selected AI Project
πͺ MirrorScope β GIST AI Hackathon π₯ 1st Place
Not every good AI product needs a complicated model. Good problem definition matters more.
Identified confirmation bias / echo chamber in YouTube comment consumption
Built a Chrome Extension providing key claims, counter-perspectives, and related news
Led service planning + AI/backend development
Focused on simple UX and immediate user value
Awarded Dean of GIST AI Graduate School Award (1st Place)
β Engineering Experience
Republic of Korea Navy β SW & AI Developer
Developed and maintained the Navy & Marine Corps Risk Assessment System
Worked in a secure closed-network environment
Reverse-engineered and maintained a large legacy system
Redesigned database structures based on field-user requirements
Developed, deployed, and operated new features
Received a Colonel's Commendation for information-system advancement and accident prevention
π§° Tech Stack
Robotics / AI
Simulation / Systems
π Honors & Awards
| π₯ | 1st Place β GIST AI Hackathon Dean of GIST AI Graduate School Award |
| π | Colonel's Commendation β Republic of Korea Navy Contribution to information-system advancement & accident prevention |
| π | Academic Excellence Scholarship Chonnam National University |
π Education
Chonnam National University B.S. in Software Engineering Double Major in Robotics Engineering
Exchange / Visiting:
KAIST
Seoul National University
π― What I Want to Build
Physical AI systems that remain reliable when objects, lighting, layouts, and task conditions change.
Perception β Simulation β Manipulation β Action Intelligence