Learning Category-level Last-meter Navigation from RGB Demonstrations of a Single Instance
Tzu-Hsien Lee, Fidan Mahmudova, Karthik Desingh.
arXiv preprint arXiv:2512.11173, 2025.
CV
Education
- M.S. in Robotics, University of Minnesota, 2026 (expected)
- B.S. in Mechanical Engineering, National Yang Ming Chiao Tung University, 2022
Experience
- Research Graduate Student, University of Minnesota | 09.2024 - Present
- Advisor: Prof. Karthik Desingh
- Robotics Perception and Manipulation Lab
- Led the system design, real-robot data collection, and experimental evaluation for a research project resulting in an arXiv preprint on category-level visual navigation for mobile manipulation.
- Built and deployed pipelines for constructing and navigating semantically enhanced topological maps using foundation models, enabling spatially-aware planning and downstream manipulation on Boston Dynamics Spot robot.
- Created SpotStack, an open-source utility repository to streamline interface development and programming workflows for Boston Dynamics Spot robot, accelerating research iteration across the lab.
- Teaching Assistant, University of Minnesota | 09.2025 - 01.2026
- Course: CSCI 5561 Computer Vision
- Designed and maintained an automated grading system for coding assignments, enabling scalable and consistent evaluation of student submissions.
- Held weekly office hours and provided technical guidance on computer vision algorithms, deep learning models, and course projects.
- Contributed to midterm exam design and grading, and managed overall course grading logistics.
Selected Student Feedback "Ben's patience during office hours and detailed guidance on assignments helped me understand not just the 'how,' but also the 'why' behind each method."
"Ben should be acknowledged by the department for his commitment to his students' learning, and I would like to have him as my TA for another course!"
"Tzu-Hsein Lee, thank you for the fact you were a nice person to talk to."
View Full Student Feedback (Official SRT Results):
- Intern, Industrial Technology Research Institute | 11.2022 - 02.2023
- Established an autonomous grinding line for irregularly shaped water faucets by independently designing force-adaptive trajectories using dual-robot arms and a force feedback sensor.
- Improved on-site calibration efficiency by designing a coordinate transformation algorithm for adaptive trajectory tuning.
- Software Lead, NYCU Autonomous Underwater Vehicle (AUV) Team | 02.2021 - 09.2022
- Spearheaded the design and implementation of the software architecture, control system, and navigation.
- Established a modular high-level/low-level robotics system by using an embedded system (STM32) for low-level attitude and motion control, and deploying ROS on Raspberry Pi and Jetson Nano for task planning and navigation.
- Reduced high-frequency noise in attitude estimation by 10\% through implementing a gradient descent-based Madgwick filter on embedded hardware.
- Developed a real-time obstacle-aware navigation system by employing A* algorithm and sonar-based localization.
Skills
- Core Competencies: Mobile Manipulation, Imitation Learning, Robot Perception, SLAM, Motion Planning, Embedded Systems
- Programming Languages: Python, C++, C, JavaScript
- Frameworks & Libraries: PyTorch, ROS (Robot Operating System), Qt, MySQL
- Tools & Platforms: Docker, Git, Linux, Gazebo, SolidWorks, Boston Dynamics Spot, STM32
Publications
Presentation
Learning Object-Centric Local Navigation from RGB Demonstrations
Poster Presentation at 2025 Midwest Robotics Workshop (MWRW), TTIC, Chicago, IL
RGB-D Networks
Student Lecture at CSCI 5980: Deep Learning for Robot Manipulation, University of Minnesota, Minneapolis, MN