location Seoul, South Korea status open to research collaborations
Sumin Lee

Sumin Lee

Computer Vision & Autonomous Driving Researcher

Teaching machines to see — and drive.

Intelligent Robotics & Computer Vision Lab · Hanyang University

I’m a Master’s researcher at the Intelligent Robotics & Computer Vision Lab, Hanyang University, working on vision-centric autonomous driving.

My current research focuses on Gaussian Splatting for autonomous driving and Online Mapping — asking how modern 3D scene representations can reshape the next generation of perception and planning.

research interests Gaussian Splatting for Autonomous Driving Online Mapping 3D Scene Representation End-to-End Driving Multi-modal Perception
01

Demo

3D Gaussian Splatting
drag · zoom · explore
02

News

  • 2026·05 submission NeurIPS 2026 — first-author paper submitted · under review
  • 2026·03 submission ECCV 2026 — first-author paper submitted · under review
  • 2025·06 award 3rd place at the Waymo Open Dataset Challenge — Vision-based End-to-End Driving (CVPR 2025)
  • 2025·03 milestone Began MS at Hanyang University, Dept. of Automotive Engineering
03

Publications

ECCV 2026 2026

Title withheld — under double-blind review

Gaussian Splatting · Autonomous Driving

First Author under review
NeurIPS 2026 2026

Title withheld — under double-blind review

Online Mapping · 3D Perception

First Author under review

Titles withheld in compliance with double-blind review.

04

Selected Projects

Swin-Trajectory 3rd Place

Swin-Trajectory

Vision-based End-to-End Driving · Waymo Open Dataset Challenge (CVPR 2025)

A minimalist transformer-based waypoint planner that takes a single front-facing camera and structured ego-vehicle history as input. A lightweight Swin Transformer backbone cross-attends to dynamic waypoint queries to efficiently regress future trajectories — competitive with much heavier multi-sensor baselines.

Deep thanks to my teammates and supervisor — and to RideFlux for technical support and collaboration.

E2E Autonomous DrivingVision-based Planning
Autonomous Valet Parking System Top 3 / 30 · Solo Award

Autonomous Valet Parking System

Capstone Design Project — full-stack autonomy

End-to-end pipeline — Perception → Planning → Control — for a self-parking vehicle, built solo. Integrated BEV-based perception, motion planning, and an adaptive pure-pursuit controller into a single coherent system. The cornerstone of my academic path into autonomy.

Valet ParkingSemantic SegmentationRGB-D CalibrationAdaptive Pure-pursuit
05

Experience

  1. 2025·03 — Present

    MS Researcher

    Intelligent Robotics & Computer Vision Lab, Hanyang University

    • Researching Gaussian Splatting for autonomous driving and Online Mapping
    • First-author submissions to ECCV 2026 & NeurIPS 2026
  2. 2023·09 — 2025·02

    Undergraduate Researcher

    Intelligent Robotics & Computer Vision Lab, Hanyang University

    • Started with path planning, expanded into multi-sensor perception (camera–LiDAR)
    • Pivoted toward vision-centric end-to-end planning
06

Education

2019·03
2022·03
2025·03

hover or focus a node for details

06

Stack

languages

Python
CUDA
C++
C

frameworks

PyTorch
ROS
OpenCV
Open3D

systems

Linux
Docker
CMake
Git
07

Contact

Let's build the next frontier of autonomy together.

Open to research collaborations, internships and discussions on Gaussian Splatting, Online Mapping, and end-to-end driving.