- 👋 Hi, I’m @JoeChan-929
- 👀 I’m interested in embodied intelligence
- 🌱 I’m currently learning ReKep and RDT-1B and some diffusion policy relevant works.
- 💞️ I’m looking to collaborate on math and machine learning
- 📫 How to reach me [email protected]
- 😄 Pronouns: Joe
- ⚡ Fun fact: I have a Roland CUBE Amplifier and I sing with my firends every Friday nights.
- South China University of Technology(Expected July 2026)
- Bachelor of Engineering in Robotic Engineering
- Relevant Coursework: Data Structures and Algorithms, Python Programming Basis, C++ Programming Basis, Machine Vision and Sensing System , Artificial Intelligence and its Application, Modern Robotics/Robotic Manipulation(ME 449 Northwestern University), Introduction of Computer Vision(CS231N Stanford University)
- Reproduced Rekep (https://github.com/huangwl18/ReKep) in simulator on laptop(Win11 RTX3060laptop).Solved environment management and file management problems, solved package dependency problems and tested prompt generalization (poor).
- Reproduced Rekep in simulator on PC(Ubuntu20.04 RTX2080ti).Solved the incompatibility problem after Isaac sim's update.
- Decoupled the code for generating key points in Rekep from the simulation environment, and deployed sam2 to generate key points for real RGB depth images.
- Investigating papers in the field of embodied intelligence and gain help from many predecessors in this field.
- Carried out the Vision grasping project and ABB robot arm safety door lock project.
- During the Vision grasping project, proposed a new approach called Continual Learning, to enhance the flexibility and Memory Efficiency of training process and simplify the operation of the B-end users during the fine-tuning process.
- Worked on the Power App Workflow: Didactic GPrC inquiry project. The understanding of IT infrastructure and workflows enabled me to optimize the project and streamline communication processes across departments.
- Basically studied about Computer Vision and done an Traffic Sign Recognition project through group work.
- Adopted HOG (Histogram of Oriented Gradients) for annotating the dataset and utilized the SVC weight file to train our custom model.
- Implemented Gaussian filtering in the data preprocessing stage, the accuracy increased from 99.3% to 99.5%
- Basically studied about the component of autonomous driving cars,had a comprehensive insight into the market and the trendy idea of car production.
- Learned about A*, Q-learning, SLAM, Pure Pursuit, Stanley Method, FSSIM, ROS,which are the base of autonomous driving.
- Built a virtual environment with Linux system and tried numbers of algorithms in the system including A*,Q-learning,SLAM,Pure Pursuit,Stanley Method to improve the decision planning and motion control.
- Adopted labelImg to label the data and used Yolov5s weight file to train our own model
- Used Mosaic data enhancement in the input side for data preprocessing, and operations such as random scaling, random cropping and random arrangement were adopted for the data set to improve the complexity of the data set.