Undergraduate Researcher in Robot Learning · Electrical Engineering @ Hanyang University

Start Fearlessly. Build Steadily. Finish Fully.

• 🎓 Education: B.S. in Electrical Engineering, Hanyang University (expected Feb 2027) · GPA 4.32/4.5 · cumulative rank 2/75
• 🔬 Affiliation: Undergraduate Researcher @ COLAB (Control & Optimization Lab)
• 🔭 Research Focus: Policy learning for robotic manipulation: Diffusion Policy, World Models, VLA, Imitation & Reinforcement Learning

• 🎤 Presenting our ICCAS 2026 paper in an oral session (Oct 28, 2026)
• 🏭 Team WIC at KETI's 2026 Robot Manufacturing-Process Competition (selected): a GR00T VLA teacher distilled into a lightweight student policy for bin-picking assembly on a Rainbow RB10

• Enhancing Visuomotor Manipulation Control with a Predictive-Dynamics-Aware World-Model-Conditioned Diffusion Policy
  • Sangho An†, Geunha Kim†, Jun Moon* (†equal contribution) · ICCAS 2026, accepted for oral presentation
  • Conditioned Diffusion Policy on a lightweight GRU world-model latent: real-robot pick-and-place success 45.0% → 72.5% on a Franka Panda, with no extra training data
  • 🌐 Project page | 📺 Experiment video

• GR00T VLA Teacher–Student Policy for Bin-Picking Assembly (in progress)

  • KETI 2026 Robot Manufacturing-Process Competition: pick randomly piled peg-in-hole parts and insert them into a fixture.
  • Proposed the GR00T teacher–student architecture; responsible for training, code verification, and data collection.
  • NVIDIA GR00T Isaac Sim Isaac Lab Rainbow RB10 RealSense D415
  • 🌐 Project page

• MACS: Magnetic AGV Contactless-charging System

  • 🏆 2nd of 13 teams, Energy Up! 2026 · 🏆 President of KEEA Award (4th of 35), 2026 KIPE IE Competition
  • Waypoint-based dynamic wireless charging for indoor logistics AGVs (82–85 kHz WPT); 79.96% end-to-end efficiency at 5 mm aligned gap, up to 85.72%.
  • WPT ROS Raspberry Pi Arduino Fusion 360
  • 🌐 Project page

• MATLAB Simulink Student Challenge 2025: Efficient Fault Recovery

  • DDPG controller with curriculum learning for quadruped actuator fault recovery: 4.6× faster training (3,900 → 850 episodes).
  • MATLAB Simulink Simscape Multibody RL
  • 🌐 Project page | 📺 YouTube

• EIGENROOM: an interactive control-engineering lab. Tune PID, LQR, and Kalman filters on a DC motor and cart-pole in the browser and see why systems respond as they do.
• Industrial Manufacturing AI Agent · 🏆 Excellence Award, Hyundai NGV × Hanyang University
  • Vision-LLM agent that classifies semiconductor defects and explains each decision; OpenCV edge detection for fine defects.
  • LangChain OpenCV FastAPI React GPT-4o-mini · 💻 GitHub
• Small web tools: MEOW PDF Merger (merge one-sided scans into a double-sided PDF), EXIF Framer, Notion2Blogger
  • Next.js TypeScript Python Vercel Supabase

• 📝 eigenroom.com/blog: English series such as Diffusion from Zero and Coding in the Age of AI
• 🇰🇷 anhoya-lab.tistory.com: the Korean originals, plus paper readings on RL, robotics, and control
• 💻 Problem solving: Baekjoon Silver I