Mechanical engineering (B.S.) and about five years of Unity/C# software development.
I connect robot hardware, navigation and task execution, then check the results against real-robot runs and recorded logs.

🎓 Physical AI Semi-Humanoid Engineer Program, Dongguk AI Campus (656 hours, July 6 to November 4, 2026, in progress)
🚗 AMR: ESP32 firmware, C++ driver, Cartographer and Nav2. Completed 20 goals on a ~77 m planned route; verified same-goal command resumption after obstacle stops. Depth-camera tabletop-edge alignment and slow approach parked the four-wheel AMR 1 cm from the tabletop edge. Separate two-wheel dock-return trials confirmed nine arrivals.
🥤 Dual-arm serving robot (team): integrated the robot URDF, restaurant collision scene, simulator-coordinate A* and task stages in Isaac Sim. Linked web orders/cancellation via ROS 2 Action and process-separated IPC; verified a nine-stage run and stopping after cancellation. Assessed completion, water loss and placement separately.
🐾 Isaac Lab / Go2: compared four reward conditions over three seeds at fixed budgets and commands. Measured command tracking and action variation apart from rewards; kept the action-change penalty after removal worsened both. Simulation only.
🤖 Unitree G1 humanoid: trained a simulated walking policy, evaluated it separately from the ground-truth-pose tracker, and linked execution via Isaac Sim / ROS 2 Bridge.
🔧 Merged contributions: Google DeepMind / MuJoCo Menagerie · ROS 2 / rosbag2 · Gazebo Gz Rendering · Navigation2
🏆 NVIDIA Korea Agentic AI Hackathon 2026: Top 5 of 10 final-round teams (Team Mollet). Qualifier: RL Triage Agent, Isaac Lab RL diagnostics with NVIDIA's Nemotron LLM, NeMo Agent Toolkit agent framework and OpenShell sandbox. Final: a separate on-site mission.
🧪 BreakTok: IMU device for detecting scrolling rhythms (TOYTHON 2026, top-20 final)
🕹️ Earlier work: built and ran four mobile games and 80 WebGL science simulations; aligned behavior/data contracts with planning/design/3D/web teams. Python + Unity CLI parallel builds cut build time ~80%.
🌱 Learning and experiment notes: hello, robot