About Me
Follow your heart and Pursue your dream
- Name: 柯于婷 Yu-Ting Ko
- Birthday: 1999/09/25
- City: Hsinchu, Taiwan
- Degree: Master
- Website: https://yutingk.github.io/
- Email: s62qtfdb9u@gmail.com
- Research Interest: Robotics, Computer Vision, Deep Learning
Skills
Programming
Python
C / C++
MATLAB
Assembly Lauguage
Middleware & Libraries & others
Robotic Operating System (ROS), Gazebo, Pytroch, OpenCV
Embedded Devices and Others
Arduino, Raspberry pi(3B/3B+), NVIDIA Xavier NX, NVIDIA Jetson Nano, TI TMS320F-28335, PIC18
Software
CCS, MPLAB, Matlab, µVision, Arduino, Hspice, Quartus
Sensors
Velodyne, Depth camera (D435), ZED
Developer Tools
Docker, Git
UI Design
Android Studio, Qt
Others
Solidworks, 3D Printing
My Resume
Summary
Professional Skills and Abilities
- Proficient in Python and C/C++, with experience in MATLAB and Assembly Language.
- Skilled in using Robotic Operating System (ROS), Gazebo, PyTorch, OpenCV, Docker, Git.
- Practical experience with Arduino, Raspberry Pi, NVIDIA Xavier NX, and NVIDIA Jetson Nano, TI TMS320F-28335, PIC18.
Academic and Research Experience
- Focused on research in Robotics, Computer Vision, and Deep Learning.
- Authored several papers published in international journals and conferences on topics including reinforcement learning, etc.
Practical Experience and Projects
- Participated in the Military Operation Research and Model Simulation Forum 20th, building a UGV team and secured the 1st place.
- Participated in the Maritime RobotX 2022 Challenge, building a heterogeneous USV and UAV team and secured the 3rd place.
- Involved in multiple projects related to robot navigation, vision systems, and deep learning.
Education
Institute of Electrical and Control Engineering, Master
2022 - 2024
National Yang Ming Chiao Tung University
Electrical Engineering, Bachelor
2018 - 2022
Chang Gung University
High School
2015 - 2018
Taichung Girl Senior High School
Professional Experience
2024
Master Thesis: 基於數位雙生架構與擴增虛擬實境互動之水域模擬器與實體機器人同步整合 Bridging Maritime Simulator and Real-world Robots using Digital Twin Framework and Augmented Virtual Reality-based Interactions
The widespread use of heterogeneous marine robots in scientific research and ocean exploration highlights the need for safer experimental environments. Digital twin technology bridges the gap between virtual simulations and real-world complexities, offering a cost-effective and risk-reducing solution for marine robotics. This approach enables realistic vehicle dynamics simulation and simpler control methods, leveraging Unity and VR for enhanced operator interaction.
2024
Competition: 第20屆軍事作業研究與模式模擬論壇 地面無人載具競賽
1st out of 9 teams. We built a UGV team to solve tasks including autonomous navigation, obstacle avoidance.
2023
Projects: Efficient Embodied Transfer of Robot Navigation using Deep Reinforcement Learning
Deep reinforcement learning (RL) enables robots with advanced navigation capabilities but is limited by the need for extensive data collection. Transfer learning offers a solution by allowing robots to apply prior knowledge to new environments, speeding up the adaptation process. Our study developed a generalized deep RL policy that excels in various challenging scenarios, proving effective in both simulations and real-world applications.
2022-2023
Related Research Experience: Utilized Unity to collect virtual datasets, providing data for deep learning models, and successfully used in training Mask R-CNN, EfficientDet, and DOPE models. And successfully validated in real-world environmental experiments.
2022
Competition: Maritime RobotX 2022 Challenge Robotx2022
3rd out of 20 teams. We built a heterogeneous USV and UAV team to solve multiple tasks including autonomous docking, navigation, obstacle avoidance, UAV launch & recovery, scan the code, racquetball flinging and acoustic pinging.
2020-2021
University Projects: Research and Application of Precision Landing System for Unmanned Aerial Vehicles Using Image Recognition Systems
以影像辨識系統建構無人飛行器精準降落系統之研究與應用
Leveraging a Raspberry Pi paired with a camera, we've successfully mitigated the inaccuracies caused by GPS positioning to under 1 meter. This advancement has been applied to drones, enhancing their precision for applications requiring exactitude in positioning.
2020/07-2020/08
University Projects: Principles and Application Experiments of TI eZdsp TMS320F28335
TI eZdsp TMS320F28335 原理與應用實驗
This project investigates various functions and designs experiments and programs in a modular fashion for each function as its foundational application.
2020
Competition: AIdea Mango Defect Classification Contest愛文芒果不良品分類競賽
Competition Ranking: 53/222, results exceeding the baseline.
Projects
Transfer RL (2023)
Deep reinforcement learning (RL) enables robots with advanced navigation capabilities but is limited by the need for extensive data collection. Transfer learning offers a solution by allowing robots to apply prior knowledge to new environments, speeding up the adaptation process. Our study developed a generalized deep RL policy that excels in various challenging scenarios, proving effective in both simulations and real-world applications.
Curriculum Reinforcement Learning (2022-2023)
Curriculum learning enhances learning efficiency by ordering tasks from simple to complex, but accurately ranking these tasks is challenging in varied domains. In deep reinforcement learning (RL) for navigation, especially among movable obstacles (NAMO), this method helps overcome the limitations of traditional navigation systems. Our research successfully developed and tested advanced DRL policies using curriculum learning, demonstrating improved navigation through environments with dynamic obstacles in both simulations and real-world scenarios.
Fed-HANet (2022-2023)
In response to the challenges of developing visual grasping algorithms for service robots, including dataset compilation difficulties and privacy concerns, we developed Fed-HANet. This model uses federated learning for a privacy-conscious, 6-DoF visual grasping system that efficiently handles unseen objects, showing comparable accuracy to traditional, less private methods. Validated through comparative analysis, depth-focused studies, and a user study with 12 participants, Fed-HANet promises to advance service robots' capabilities while maintaining user privacy.
Maritime RobotX Challenge (2022)
An autonomous robotic systems operating in the maritime domain competition
We built a heterogeneous USV and UAV team to solve multiple tasks including autonomous docking, navigation, obstacle avoidance, UAV launch & recovery, scan the code, racquetball flinging and acoustic pinging.
Publications
“Fed-HANet: Federated Visual Grasping Learning for Human Robot Handovers”
[ paper] [Project] Submitted to IEEE International Conference on Intelligent Robots and Systems (IROS). 2023
“Curriculum Reinforcement Learning from Avoiding Collisions to Navigating among Movable Obstacles in Diverse Environments” (2023) IEEE Robotics and Automation Letters (RA-L), 8(5), 2740-2747
[ Paper] [Project] Submitted to IEEE International Conference on Intelligent Robots and Systems (IROS). 2023
“A Learning-based Modular Heterogeneous USV and UAV Team in the Maritime RobotX 2022 Competition” (2022)
[Paper] [Video] Maritime RobotX 2022 Competition Technical Design Paper
“Efficient Embodied Transfer of Robot Navigation using Deep Reinforcement Learning” (2023)
[Project] [Video]
Contact Me
Social Profiles
Email Me
s62qtfdb9u@gmail.com