Ulsan National Institute of Science and Technology (UNIST)
Integrated M.S.–Ph.D. Program in Artificial Intelligence · 2021–2028 (Expected)
3D Vision & Robotics Lab · Advisor: Kyungdon Joo
Inha Lee 이인하
Integrated M.S.–Ph.D. student · UNIST
SLAM · Geometric foundation models · Neural 3D reconstruction · Collaborative perception
Education & Research Affiliation
Chungbuk National University
B.S. in Electronic Engineering · 2017–2021
Research Interests
SLAM · Multi-Agent Systems · Lifelong SLAM · Active Perception
Ongoing Research & Projects
Long-Horizon Spatial Memory for Multi-Robot Geometric Foundation Models
Researching shared spatial memory for multi-robot geometric foundation models, with revisit-driven updates for implicit loop closure and efficient map representations.
Lifelong 3D Map Management with Multi-Session Scene Change Detection
Researching real-time, multi-session scene change detection for lifelong 3D map management as environments evolve.
AI Bots Collaboration Platform and Self-Organizing AI
IITP · RS-2022-II220907
Developing a robust decentralized multi-agent SLAM system designed to operate reliably under diverse real-world noise and uncertainties.
Selected Research & Publications
* Equal contribution. Author order follows the original publications.
UniSim-SLAM: Feed-Forward SLAM with Unified Sim(3) Optimization
European Conference on Computer Vision (ECCV), 2026First author
Combines fast two-view tracking with multi-view refinement through a unified Sim(3) factor graph for geometrically consistent feed-forward SLAM.
Technologies: Python
FeDepth: Federated Learning for Depth Estimation under Robot Heterogeneity
European Conference on Computer Vision (ECCV), 2026Co-first author
Uses descriptor-based soft clustering to learn depth estimation across heterogeneous robots, environments, and sensing conditions without transferring raw data.
Technologies: Python
A Benchmark Dataset for Collaborative SLAM in Service Environments
IEEE Robotics and Automation Letters (RA-L), 2024Project lead
Presented at ICRA 2025
A synchronized multi-robot benchmark built in NVIDIA Isaac Sim, with three robots in hospital, office, and warehouse environments and paired static/dynamic sequences.
Technologies: Python · ROS · NVIDIA Isaac Sim
AiSDF: Structure-aware Neural Signed Distance Fields in Indoor Scenes
IEEE Robotics and Automation Letters (RA-L), 2024Co-first author
Presented at ICRA 2025
Reconstructs indoor scenes online by combining neural signed distance fields with Atlanta-world structural priors and explicit planar surfels.
Technologies: Python
Fracture Assembly with Segmentation and Iterative Registration
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024
Reassembles fractured 3D objects by segmenting broken surfaces and aligning fragments through iterative point-cloud registration.
Technologies: Python
Undergraduate Engineering Projects
Autonomous Driving: Path Planning & Control
Led team development of path planning and vehicle control for the 2020 university creative-car competition. Integrated global reference paths prepared in QGIS, Pure Pursuit tracking, parking, static-obstacle avoidance with a local grid map and candidate paths, emergency braking, and traffic-light handling.
Technologies: C++ · Python · ROS · Vehicle control · Local planning · Global path planning · Pure Pursuit
Cooperative Driving at Intersections
Developed cooperative intersection driving using vehicle-to-infrastructure (V2I) communication and camera perception. Localized vehicles with ArUco markers relative to a traffic-light camera and demonstrated collision avoidance in red-light-violation scenarios.
Technologies: ROS · V2I communication · Computer vision · ArUco marker localization
Indoor Delivery Robot
Built an indoor delivery robot using Gmapping to create maps, AMCL for localization, and a DWA local planner for navigation.
Technologies: ROS · Gmapping · AMCL · DWA local planning
RANSAC-based Lane Detection
Developed a ROS vision pipeline that transforms camera images into a bird’s-eye view, parameterizes lane lines in cylindrical coordinates, detects white and yellow lanes with sliding-window RANSAC, and generates waypoints for lane following.
Technologies: C++ · ROS · Computer vision · RANSAC
Balancing Robot
Built a self-balancing robot that computes a balance error from a six-axis motion sensor and uses PID feedback to control stepper motors through an ARM Cortex-M microcontroller.
Technologies: C++ · ARM Cortex-M · PID control · Six-axis motion sensing · Stepper motors
High-Speed Line Tracer
Built an STM32 / ARM Cortex-M line-following robot with hand-built hardware, infrared sensors, stepper motors, and PID steering and speed control. The project reports running speeds of 1.5–2.4 m/s.
Technologies: C (embedded) · ARM Cortex-M · STM32 · Infrared sensing · PID control · Hand-built hardware
Hand-Tracking Rhythm Game
A camera-based rhythm game controlled by hand motion. Color segmentation and histogram backprojection detect the hands, while Mean Shift tracks them for interactive gameplay.
Technologies: C++ · OpenCV · MFC · Mean Shift tracking