Ji Sue Lee

I'm a candidate Ph.D at University of Hanyang in South Korea and studying Electrical Engineering.

I'm passionate about exploration and exploitation challenges in Reinforcement Learning (RL), Meta-RL, and Causal-RL. I'm currently researching how to optimize Meta-RL with Causal Representation Learning in mobile robot scenarios.

Email  /  Scholar / Twitter  /  Github / LinkedIn

profile photo

Publications
GOME-NGU preview GOME-NGU detail
GOME-NGU: visual navigation under sparse rewrd via Goal-Oriented Memory Encoder with Never Give Up
Ji Sue Lee, Jun Moon
IEEE Access, 2025
project page / video / bibtex

In this paper, we propose the Goal-Oriented Memory Encoder (GOME) with Never Give Up (NGU) algorithm to enhance visual navigation in sparse reward environments.

Miscellanea
Reinforcement Learning Basics with JAX (March. 2025 - TBD)
This project serves as a comprehensive guide for learning and implementing reinforcement learning (RL) algorithms using JAX. It is designed for both researchers and practitioners, providing structured tutorials and hands-on code implementations.
  • Covers fundamental RL topics such as Markov Decision Processes and Q-learning.
  • Explores advanced methods, including policy gradient techniques and model-based RL.
  • Provides a structured 15-week curriculum with practical exercises and real-world applications.
  • Includes video tutorials and hands-on projects to reinforce learning.
Paper Reviews (April. 2025 - TBD)
Paper Reviews is a curated platform offering in-depth reviews of cutting-edge research in AI, Machine Learning, and Robotics. Designed for both researchers and practitioners, it provides comprehensive insights, accessible explanations, and regular updates on influential publications.
  • Curated collection of influential research papers across diverse AI domains.
  • Detailed reviews and technical analyses that simplify complex concepts.
  • Regular updates featuring featured papers and new research developments.
  • Community-driven content tailored for both experts and newcomers.

Projects
NVIDIA Isaac-Sim Mobile Robot Simulator (May. 2024 - TBD)
We are proceeding with the NVIDIA Isaac-Sim Mobile Robot (Franka Research 3, Husky, QCar, Husky + Franka Research 3) unified platform to ease usage for beginner to master courses.
  • We are developing a new user interface (UI) in the NVIDIA Isaac-Sim.
  • In our implemented UI, we demonstrated several mobile robots and environments.
  • We demostrated a human simulator and sensor applications (stereo camera, LiDAR).
  • We demonstated simple ablation scenarios (SLAM, path-planning) for mobile robots.
CARLA Semantic Segmentation Challenge (May. 2023 - May. 2023)
Creating a semantic segmentation model for autonomous driving in the CARLA environment using a pre-trained model.
  • Achieved real-time segmentation under diverse weather conditions.
  • Utilized pre-trained DeepLabv3 models and fine-tuned architectures (U-Net, Mask-RCNN).
  • Optimized class definitions by remapping 28 classes into 12 key categories.
  • Enhanced scene understanding for autonomous driving applications.
HyperDrive: Advanced Reinforcement Learning for Autonomous Racing (TBD)
A research-driven project exploring advanced reinforcement learning techniques for autonomous racing, conducted using the Learn to Race simulation platform.
  • Developed and evaluated various reinforcement learning algorithms, including Online RL, Offline RL, Transformer-based RL, State Space Models, and Meta-RL.
  • Designed a benchmarking framework to compare algorithm performance in terms of speed, generalization, and sample efficiency across multiple racing tracks.
  • Implemented policy optimization techniques to enhance driving stability and reduce lap times in high-speed racing scenarios.
  • Conducted extensive experiments using large-scale GPU clusters, optimizing computational efficiency for real-time decision-making in autonomous driving.

Updated March 2025.

The source code of this website is owned by Jon Barron.