|
Dong-Won Lee I'm an undergrad student in Software Convergence at Kyung Hee University. Currently I'm focused on reinforcement learning and robot learning for physical AI, building on a background in 3D perception, 3D Gaussian Splatting, and SLAM. Previously I worked as a Software Engineer Intern at ROBROS, developing camera-streaming software for a humanoid robotics SDK, and as a Research Intern at AIMS Lab, Kyung Hee University. I also served for a year and a half in the military as a sergeant. |
|
Research InterestsI am interested in reinforcement learning, robot learning, and embodied AI: how agents learn reliable behavior from experience, data, and interaction with the physical world. I see perception and control as two halves of learning well from limited data, and I aim to connect spatial grounding from 3DGS/SLAM with policy learning for robotics, VLA systems, and physical AI. |
Experience |
|
|
ROBROS Developed a Python/C++ camera-streaming pipeline for a humanoid robotics SDK and validated it on a physical humanoid robot for VR-based teleoperation. |
|
AIMS Lab, Kyung Hee University Conducted hypothesis-driven analysis of LLM use in elementary classrooms and helped construct and validate a multimodal classroom-interaction dataset. |
|
Photo Studio Booking Platform Led a five-member team building an iOS studio-booking application, covering product development, design, performance marketing, and B2B studio partnerships. |
Projects |
|
|
Data-Efficient 3D Gaussian Splatting Applying dataset distillation to input-frame selection for 3D Gaussian Splatting, targeting data-efficient novel-view synthesis under sparse views with gsplat and Mip-NeRF 360. |
|
Studying policy optimization, exploration, and offline RL while implementing PPO and TD3+BC in PyTorch on MuJoCo continuous-control benchmarks. |
|
Local LLM Infrastructure for Agentic AI Coding Building a self-hosted GPU-based LLM environment for coding agents with vLLM on NVIDIA DGX Spark, including KV-cache quantization and prefix caching. |
|
Design and source code from Jon Barron's website. |