Junhyeok Lee

B.S. in Automotive Engineering Kookmin University, Senior
LinkedIn Instagram
About

About Me

I am a senior studying Automotive Engineering at Kookmin University with a minor in Artificial Intelligence. My research interests lie in reinforcement learning, end-to-end autonomous driving, and motion prediction & planning.

I build real systems — from RL-based vehicle controllers to multi-agent AI trading platforms — always bridging the gap between research and deployment.

Experience
Jun 2025 — Present
Unity Project — Self-Driving Simulation
Building an RL-oriented self-driving simulator in Unity. Implementing vehicle control (WheelCollider, C#), Blender-modeled tracks, and camera/LiDAR sensor pipelines. Developing a high-throughput simulator–host communication layer for training scenario-specific driving policies.
Jun 2024 — Nov 2025
KUUVe — Control Team & Team Leader
Kookmin Univ. Unmanned Vehicle
Developed localization algorithms using odometry and GPS, path-tracking systems, and RL algorithms with Sim-to-Real transfer on JetRacer. Led two competition teams coordinating perception, planning, control, and systems teams.
Nov 2023 — May 2024
KaAI — AI Research Member
Kookmin Automotive Artificial Intelligence
Participated in dataset post-processing for autonomous driving research, performing segmentation tasks on camera and LiDAR data. Gained foundational knowledge in deep learning through team study sessions.
Mar 2021 — Present
B.S. in Automotive Engineering
Kookmin University · Minor in AI · GPA 3.99/4.5
Comprehensive education in mechanical system design, electronic control, vehicle dynamics. Coursework in Reinforcement Learning, Automatic Driving Computing, AI Accelerator Design.
Awards
Sep 2025
HL FMA aMAP Innovator Championship
GPS-based autonomous parking system with LiDAR obstacle detection
Encouragement Award
May 2025
Autonomous Driving Robot Race
GPS-based localization, Pure Pursuit–Stanley hybrid controller with adaptive lookahead
Excellence Award
Dec 2024
AutoRace — COSS
Rule-based decision-making without GPS/SLAM using LiDAR and camera perception
Merit Award
Oct 2024
International Robot Contest & R-BIZ Challenge
Odometry-based localization with Kalman Filter sensor fusion for enhanced accuracy
Excellence Award
Publications
Performance Enhancement of Lane Following in Autonomous Driving using Sequential Multi-Critic PPO with Critic Transfer Strategy
Junhyeok Lee, et al.
KSAE Annual Fall Conference, Nov 2025 (Accepted)
Improved Path-Tracking Performance and Driving Stability of Nonlinear MPC Using Variable Steering-Angle Weights Based on Arctangent Function
Juho Yun, Jinsung Kim, Yeonseo Park, Junhyeok Lee, et al.
KSAE Annual Spring Conference, May 2025
Skills
01

Autonomous Driving

Perception, planning, control. Localization with Kalman Filter. Sim-to-Real reinforcement learning on JetRacer and ERP-42.

02

Reinforcement Learning

PPO, multi-critic architectures, critic transfer. RL-oriented driving simulation in Unity with Gymnasium integration.

03

Simulation & Synthetic Data

Custom simulator development in Unity. Diffusion models for domain adaptation. Camera and LiDAR sensor pipelines.

04

Full-Stack & AI Systems

Multi-agent AI trading with LangGraph. Next.js, FastAPI, Docker, PostgreSQL. Local GPU inference with Qwen3 30B.

Tech Stack
Python
C / C++
PyTorch
TensorFlow
TensorRT
ROS2
Unity
Docker
Git
Blender
FastAPI
Cloudflare
Projects

Selected Projects

Research, development, and engineering projects I've built.
KUUVe
Research Club · Jun 2024 — Nov 2025
Kookmin University Unmanned Vehicle — an academic R&D club for autonomous driving. As a control team member and competition team leader, I developed GPS/odometry-based localization, Pure Pursuit–Stanley hybrid path-tracking controllers, RL-based Sim-to-Real policies on JetRacer, and led integrated vehicle stacks for 4 competitions (2 Excellence Awards, 1 Merit Award, 1 Encouragement Award).
Autonomous DrivingROS2Kalman FilterPath TrackingJetRacerERP-42
Nano-sim
Simulation · Jun 2025 — Present
A custom-built RL-oriented self-driving simulator in Unity for synthetic data generation and training. Implementing vehicle dynamics (WheelCollider, C#), Blender-modeled environments, camera/LiDAR sensor pipelines, and developing diffusion models to reduce domain gap between simulated and real-world driving data.
UnityC#BlenderDiffusion ModelReinforcement LearningDomain Adaptation
Trading Agent
AI System · 2025 — Present
Building my own Berkshire Hathaway — a multi-agent AI trading system where 5 AI agents with distinct investment philosophies (value, risk, momentum, sentiment, macro) debate and execute stock trades across Korean and US markets. Powered by LangGraph + Qwen3 30B running on local GPU with Cloudflare Tunnel deployment.
LangGraphQwen3 30BKIS APIFastAPIWebSocketPostgreSQL
Blog

Blog

Not posted yet.
Contact

Get in Touch

Interested in collaboration, research, or autonomous driving — feel free to reach out.