Robotics Perception & Machine Learning Engineer
I am a robotics perception and machine learning engineer with seven years of experience developing camera-based perception systems at Hyundai Motor Company. I am currently pursuing an M.Eng. in Autonomy and Robotics at the University of Illinois Urbana-Champaign.
My work focuses on object detection and tracking, monocular 3D localization, camera geometry, motion prediction, and scalable ML data pipelines.
View Experience GitHub LinkedIn
Featured Work
Scalable Sim-to-Real Monocular 3D Object Localization
Developed camera-geometry-based monocular 3D object localization system trained exclusively on simulation-generated data, achieving 1.49 m longitudinal mean absolute error within 70 m on real-world driving data.
The project introduced geometry-driven data synthesis for diverse camera configurations, enabling labeled training-data generation without manual 3D annotation for each vehicle platform.
2.5D Object Detection and Keypoint Network
Developed YOLOv5-based vehicle perception model combining object detection, view classification, and keypoint estimation. The model achieved 0.740 mAP and 0.804 F1 score with 27.70 ms inference latency on NVIDIA T4.
Converted the model to TensorRT and integrated the perception pipeline into a ROS2-based front-camera system for on-vehicle testing.
Multi-Object Tracking and Visual Re-Identification
Developed and validated Kalman-filter-based multi-object tracking system, optimizing track management policies and data association using sequence-level ground truth and MOTA-based evaluation.
Integrated visual Re-ID features to improve MOTA from 33.13 to 40.10 while reducing computational cost by 29.1% through shared detection and Re-ID features.
Video Anonymization and ML Data Pipeline
Built active-learning video anonymization and continuous-learning pipeline using Python, PyTorch, Apache Airflow, Ray, and Kubernetes.
The pipeline improved Dice from 89.11% to 92.61% across new vehicle configurations and unseen driving environments without manual labeling.
Selected Publication
Scalable Sim-to-Real Monocular 3D Object Localization across Diverse
Sensor Configurations
IEEE International Conference on Intelligent Transportation Systems
(ITSC), 2026
Focus Areas
Computer Vision · Autonomous Driving · Object Detection · Multi-Object Tracking · 3D Object Localization · Camera Geometry · Motion Prediction · ML Systems