Turn symptoms into numbers.
"Jittery" and "slow" become standard deviations, ranges, and latency distributions. If before and after can't be compared on the same data, it isn't an improvement.
Robotics SW Engineer · Seoul
ROBOTICS · SIMULATION · PERCEPTION
I validate robots so they
I validate robots so they
work reliably
in the real world.
From Isaac Sim digital twins to ROS 2 sensor pipelines and camera·LiDAR perception algorithms. In my master's, I transferred reinforcement-learning policies to physical robots and published a first-author SCIE journal paper (Sensors, 2025). Today I build perception algorithms and simulation-based validation environments at an autonomous parking-robot company. My method: reproduce the failure, then improve it with numbers.
ENGINEERING FOCUS
Sim-to-Real
Sim-to-Real
Validation Loop
- 01MODELModel the robot, sensors, and environment in simulation
- 02CONNECTWire up ROS 2 and the perception pipeline
- 03VALIDATEReproduce failures · compare quantitatively · improve
01 / PROJECTS
Projects with clear
Projects with clear
problems and results
Industry work and master's research, side by side. Industry projects are described anonymously — no company or product details — focusing on the problem-solving process and engineering decisions. Click any card for a detail page with figures and videos.
INDUSTRYAutonomous parking robots · 2026.01 – present
INDUSTRY
2026.01 — 2026.06 · Pose-estimation and detection pipeline · Running on production robots
Dual-Marker Relative Pose Estimation — System Overhaul
Found the structural defects that silently disabled the filtering pipeline, redesigned it with Ceres non-linear optimization and quaternion-aware filters, then recovered the added compute cost with a Kalman-based dynamic ROI.
Pitch noise −59% · ~17× more detections while driving · processing time −50%
Read more →
INDUSTRY
2026.03 — present · Solo project
Isaac Sim Algorithm-Validation Environment
Rebuilding the real robot, sensors, and operating environment in simulation — a digital twin where ROS 2 perception algorithms can be tested repeatedly without site visits.
Failure cases reproduced in sim · Docker-deployed algorithms fully connected
Read more →+ SIDE WORK
Multi-camera GigE streaming bottleneck analysis · 2026.03 — 2026.04 — Diagnosed frame drops in 4-camera simultaneous streaming through four rounds of latency experiments, isolating the cause layer by layer (switch → shared PoE bandwidth → uplink) and deriving stable stream settings and a recommended network topology.
RESEARCHRobotics & Virtual Reality Lab (RVLab), Konkuk University · 2023 – 2026
RESEARCH
2025.03 — 2025.08 · Solo · First-author SCIE journal
SOAR-RL: RL Navigation in Narrow Spaces
Fused 3D LiDAR and RGB to detect and track people, then trained an open-space-aware RL policy that navigates narrow spaces safely alongside humans.
94% average success rate · 35% shorter travel time than crowd-navigation baselines · Published in Sensors
Read more →
RESEARCH
2023.09 — 2024.04 · Team of 6 · 80% contribution
Sim-to-Real Driving with Unity ML-Agents
Trained an RL driving policy in Unity, exported it to ONNX, and deployed it on a physical Husky A200 — stabilized real-world driving with LiDAR noise injection and curriculum learning.
Stable driving over a ~50 m corridor — 6.7× the distance of the first attempt
Read more →
RESEARCH
2024.07 — 2025.02 · Team of 6 · Project lead
Isaac Sim Autonomous-Driving Simulation & ROS Interface
Rebuilt a 3D-scanned lab environment in Isaac Sim and developed ROS interfaces for six LiDAR models, an IMU, and robot control. Led scheduling and module integration.
ETRI research project · 6 LiDAR models connected over ROS · public demo video
Read more →
RESEARCH
2023.05 — 2025.12 · Team of 4 · 50% · Year-2 lead
Mold-Polishing Automation — Digital Twin & Mobile Manipulator
Recreated an industrial site as a digital twin to pre-validate algorithms, and automated surface-roughness inspection with AprilTag localization and manipulator control.
Repeatability error avg 1 cm · data collection 50% faster · roughness estimation +10% accuracy
Read more →+ MORE RESEARCH
Outdoor mobile-robot autonomy experiments — GPS·IMU·camera·LiDAR data collection, 2D-LiDAR dynamic obstacle avoidance, GPS trajectory visualization.
OmniIsaacGym obstacle-avoidance path planning · BARN Challenge preparation · Jackal navigation in Docker — more research notes in the Notion archive ↗ (Korean).
02 / APPROACHDoes it work reliably
Does it work reliably
on the real robot?
Reproduce the failure conditions.
Occlusion, long range, low light, CPU load — I turn failure conditions into repeatable tests, in both real data and simulation.
Optimize accuracy and real-time together.
Instead of only pushing detection performance, I optimize the whole system budget — ROI, execution conditions, sensor configuration.
03 / PUBLICATIONS
Publications
During my master's (2024.03 – 2026.02): two SCIE journal papers (one first-author), one international conference paper, and eight domestic conference papers.
Journal
- Co-author · SCIEKim, H., Jun, M., Jung, H. — Multimodal RGB–LiDAR Fusion for Robust Drivable Area Segmentation and Mapping. Sensors 2025, 25, 5841.DOI ↗
International Conference
Domestic Conference (Korea)
- Jeong, H., Luo, C., Jun, M., Jung, H. — Image Data Collection System for Estimating Surface Roughness of Automotive Press Dies. KSME Conference, 2024.11.
- Luo, C., Jun, M., Jeong, H., Jung, H. — Video-Based Estimation Method for Mold Surface Roughness. KSME Conference, 2024.11.
- Luo, C., Jun, M., Jung, H. — AI-Vision-Based Surface Roughness Estimation for Automotive Press Dies. KSME Conference, 2024.05.
Awards & Research Grants
04 / TOOLBOX
Tech Stack
Isaac Sim / Isaac LabUnity ML-AgentsROS 1 / ROS 2
C++PythonOpenCVCeres Solver
PyTorchLiDARRGB-D CameraDockerGit
05 / EDUCATION
Education & Certification
M.S. in Mechanical, Robotics & Automotive Engineering, Konkuk University · 2024.03 – 2026.02
B.S. in Mechanical & Aerospace Engineering, Konkuk University · 2020.03 – 2024.02
OPIc IM2 · Acquired 2025.09