Weiyi Tang

Weiyi Tang

Master of Science in Electrical Engineering

University of Pennsylvania

Hi there!

I am a graduate student at University of Pennsylvania. I am currently a research assistent at Kod Lab, a subsidiary of the Penn Engineering GRASP Lab. My research interest lies in the interaction of robotics, control and vision. I am currently working on applying reinforcement learning to a single leg of a direct-drive robot like Minitaur and learning predictive models from observation and interaction. I am particularly excited about finding ways to make robots interact with the physical environment surrounding them, then solve challenging tasks automatically.

Interests

  • Robotics
  • Control
  • Motion Planning
  • Computer Vision
  • Reinforcement Learning

Education

  • MS in Electrical Engineering, 2020

    University of Pennsylvania, US

  • BS in Electrical Engineering and Automation, 2018

    Hunan University, CN

Skills

Programming

Python, C/C++, MATLAB, GAMS, Assembly language, Mathematica

Robotics

ROS, Pytorch, Tensorflow, OpenCV, OpenAI, Pandas, NumPy, Matplotlib, SolidWorks, Ubuntu

Control

Feedback Control, Model Perdictive Control, Motion Planning

Projects

*
All Control Computer Vision

Classic Computer Vision Methods

Implement Canny edge dectetion, Image Gradient Blending, Image Morphing, Seam Carving, Image Stitching and Optical Flow. (Note: Didn’t use any package for all the implementations)
Weiyi Tang, Zhiyuan Li
Classic Computer Vision Methods

Deep Learning System for Bone Age Assessment

we are going to implement a fully automated deep learning pipeline to segment a region of interest, standardize and preprocess input radiographs, and perform Bone Age Assessments. We first normalize our input images, because they have different sizes and background color. We then try different ways including Image Processing Technology, Single Shot MultiBox Detector (SSD) and Mask R-CNN to remove annotation markers and background border from image to make our data more clearly. We finally feed processed pictures to GoogLeNet to predict the age of bone. As for the result, we can get accuracy of roughly 95% if allowing error within 2 years, while roughly 80% if allowing error within 1 years.
Weiyi Tang, Tiening Li, Shiao Li, Songyue Lu
Deep Learning System for Bone Age Assessment

F1/10 Autonomous Racing

Assembly a F1/10 car and utilized C++ via ROS to code different algorithm to make car perform different behaviors.
1). Utilized simple PD controller to complete wall following.
2). Applied reactive method: follow the gap to achieve run a lap without hitting obstacles.
3). Utilized SLAM to locate car then ran car with Pure Pursuit as global planner and RRT* as local planner.
Weiyi Tang, Shivangi Misra, Dingding Zheng
F1/10 Autonomous Racing

Machine Perception

Implement machine perception technology include: logo projection, scale invariant, SIFT matching and pose estimation.
Weiyi Tang
Machine Perception

Obstacle Avoidance: Path Planning and MPC Tracking

In our project, the cart is running from a predefined source point to a destination point on a map with some fixed x and y boundaries. On the map we generate some randomly distributed obstacles with random sizes and shapes unknown to the cart. When the cart “detects” the obstacle, it will follow the path we planned to avoid collision.
Jiacan Xu, Weiyi Tang, Suyu Zhou
Obstacle Avoidance: Path Planning and MPC Tracking

ODrive Hopper Robot

Using mechanical design inspired by the Ghost Minitaur and the open-source motor controller hardware from the Stanford Doggo, we built an open-source two-degree-of-freedom hopping robot. The robot hops using a Raibert-inspired reactive controller on the leg length and velocity.
Weiyi Tang, Sonia Roberts
ODrive Hopper Robot