Dohyeong Kim

I am currently a B.S. student in Computer Science and Engineering at Konkuk University, and an Incoming Integrated B.S./M.S./Ph.D. Student at Trustworthy Machine Learning Lab, advised by Prof. Sangwoo Hong.

My research interests include Physical AI Explainability and Memory-Efficient Local Inference.

Email / Github / CV / LinkedIn

🎓 Education

Konkuk University Mar 2022 - Current
Undergraduate Student, Department of Computer Science and Engineering

🌍 Experience

Trustworthy Machine Learning Lab (TML), Konkuk University Jun 2025 - Current
Undergraduate Researcher / Incoming Integrated B.S./M.S./Ph.D. Student
Advisor: Prof. Sangwoo Hong
Working on Physical AI explainability.

Team K.A.I, Konkuk University May 2025 - Nov 2025
Perception / Autonomous Driving
Built ROS2 perception pipelines for competition vehicles, including YOLO traffic-light detection, ros2bag-based dataset construction, camera-LiDAR fusion, obstacle-distance estimation, and control-facing topic/service interfaces.

📄 Publications

Look Where You Drive: Action-Based Visual Explanation for Autonomous Driving VLAs
Dohyoung Kim, Donghyun Kim, Jaehyuk Yu, Sangwoo Hong
ECCV 2026 Submission

Proposed RHEMA, a post-hoc action-grounded visual explanation method for autonomous driving VLAs. Contributions include the y_act problem formulation, SimLingo/ReCogDrive adaptation, dual-source fusion, ablation design, and rebuttal-oriented result analysis.

🧪 Projects

YOLO Dataset Studio
Github

A CLI workspace for custom YOLO dataset bootstrapping: ROS2 bag/video ingestion, OpenCV GUI labeling, teacher/student training, auto-labeling, split/merge/sampling, and active-learning candidate selection. Used in autonomous driving competitions.

✨ Awards

Kookmin University Autonomous Driving Contest Aug 2025
1st Prize
Team K.A.I perception contribution: competition-specific YOLO dataset construction, ROS2 detection interface, center-lane tracing, camera-LiDAR fusion, DBSCAN obstacle-distance estimation, E-Stop, and ultrasonic-based overtaking logic.

2025 College Student Creative Mobility Competition Nov 2025
Excellence Award in Creative Technology · Special Award in Autonomous Driving Performance
Built a three-class traffic-light dataset and lightweight YOLO pipeline for real competition lighting/camera conditions, then passed green-light decisions to control code through ROS2 services/topics.

1st Physical AI Hackathon Feb 2026
2nd Prize, Fine Folding Track
Connected local smolVLA inference with real dual-arm manipulation through data collection, calibration, and action execution.

Design inspired by Jon Barron.