Hello is not just a greeting. The moment you open this page, we are connected. No mater who you are or where you are, it is my honor to be known by you among 6 billion people on this planet. I don't believe you open this page randomly, because there are billions of pages being viewed and visited every second, but why me, why this one, why this moment. So just keep reading, treat this as some kind of asynchronous communication. If you want to tell me something, hit the connect button and leave me a message, then we will be strangers no more!


My Name is Yunkai Cui, you can also call me Edward

I am currently a research associate working in Modlab at University of Pennsylvania, suppervised by Dr. Mark Yim. Modlab is a subsidiary of GRASP (General Robotics, Automation, Sensing & Perception) lab at Penn. My current research includes Reinforcement Learning in robotics, self-reconfigurable modular robot simulation, tasks understanding, Multi-task learning & knowledge transfer, and many other cool stuffs.

Besides that, I'm also intrested in studying machine learning theories and apllying machine learning to solve real life problems. A couple of my friends and I are building a new type of recommendation system that will take advantage of unsupervised learning and deep learning.

I am also a full-stack web developer, photographer, robot builder, PCB designer, and open source software contributor.

e-mail: yunkai(at)seas(dot)upenn(dot)edu

  • Robots
  • Circuit
  • Design
  • Photo

Resume | A PDF version


University of Pennsylvania
2012 - present

Master of Science in Robotics, May 2014
School of Engineering and Applied Science
Master's thesis: Unsupervised learning in modular robot reconfiguration

Shanghai Jiao Tong University
2009 - 2012

Bachelor of Science in Mechanical Engineering & Automation, June 2013
Shanghai Jiao Tong University-Tyco International Scholarship


Dean's List, SEAS, Penn (2012 - 2013)

Excellent Undergraduate Student Creative Project Award, SJTU (Apr. 2012)

Shanghai Jiao Tong University-Tyco International Scholarship (Dec. 2011)

Government Award for Excellent College Student Creative Projects, Shanghai (Sept. 2011)

Excellent Volunteer and Week Star in Shanghai EXPO (Oct. 2010)

First Prize winner in 26th CPhO (Chinese Physics Olympiad) (Mar. 2009)

Gold Medal Winner, 15th Hope Cup National Mathematics Contest, China (Nov. 2005)


PennApps Top 20 Demonstration, 2013 Spring

Irvine Auditorium, University of Pennsylvania, Philadelphia, PA Jan. 2013

The hardware and software structure of the "Skynet" remote robot arena network


Research Assistant, Modlab, GRASP, Penn
​Feb. 2013 - present

Worked on general modular robot topics.
Supervised by Dr. Mark Yim

Teaching Assistant, MEAM 520, SEAS, Upenn
​Aug. 2013 - Dec. 2013

Worked as a TA for Introduction to Robotics,
Assisting Dr. Katherine Kuchenbecker

Teaching Assistant, CIS 515, SEAS, Upenn
​Aug. 2013 - Dec. 2013

Worked as a TA for Liner Algebra and Optimization,
Assisting Dr. Jean Gallier


Institute of Electrical and Electronics Engineers (IEEE) Student Member


  • C/C++
  • Python
  • JavaScript
  • Matlab
  • Latex
  • Bash Shell
  • SQL
  • PHP


  • jQuery
  • Node.js
  • OpenCV
  • Oracle
  • MySQL
  • NLTK
  • Hadoop
  • Qt
  • Solidworks
  • Altium Designer
  • Labview
  • Siemens NX
  • AutoCAD


Unsupervised Learning in Modular Robot Reconfiguration
August, 2013 - Oct, 2014

Giving modular robot an environement to execute some tasks, how could this robot system find out when and where it should reconfigure itself to adapt the local environment. If the robot knows when and where to reconfigure, could we use this information to optimize the deisgn of configurations. We showed in our paper that these can be done using various machine learning techniques. We proposed to use a clustering method to find the reconfiguration boundaries, and then use a PCA to find the similarity of the states in the same cluster to assist the configuration design.

For details, please read more!

Spectral training of Hidden Markov Model with Continous Emission
Jan. - May. , 2014

Is it safe to assume that we will always have enough data to train our Hidden Markove Models? Could the training process be accelarted by using an approximation instead of a full Expectation-Maximazation? My research focuses on answering these two questions. We first used a spectral algorithm to show that HMM training could be accelerated. We trained three moments by using statistic techniques and then used them as priors to do inference and predictoin. But this method works only when we have enough data to build those priors. We then proved that, with an appropriate dimenstion reduction method, we can reduce the data size needed for training a descent priors dramatically.

For details, please read more!

The new SMORES Modular Robot platform
From Aug 31, 2013 to present

This is a NSF funded project. I am currently cooperating with ASL lab at Cornell University. The general ideas is to design a new modular robot system, where each module has four degree of freedoms, to do something useful. We called this modular robot system 'SMORES'. I built a simulatior and a bunch of lower level libraries for designing and testing the high level planing and automy algorithms for this robot platform. For more information, please visit our project website. For old SMORES project, please visit Modlab homepage.

For details, please read more!

Occupancy Grid SLAM
March. , 2014

This is an old fashion Simultaneous localization and mapping(SLAM) implementation using Inertial measurement unit(IMU) data and LIDAR data. A particle filter was implemented to improve the position estimation(Bayesian style posterior estimation) with 200 particles. The map was generated on an occupancy grid, with 5cm resolution. In the picture above, the map on the left was generated using SLAM, while the map on the right was generated with pure Odometry data(IMU and encoder). The estimated trajectory of the mobile robot was showed in green.

For details, please read more!

Logo Replacement
On Dec 21, 2013

This is the final project of the Computer Vision class. The goal of this project is to achieve the automatic detection and replacement of embedded advertising logos in images and videos. The basic idea of our method is to use SiFT algorithm (a fancy version) to achieve original logo detection, then using TPS to do the new logo transformation, and applying first derivative gradient blending (Poisson) to blend the new logo into the scene. For this project, I invented a feature points matching method by using barycentric coordinates, which improved the matching performance of the original SiFT algorithm a lot.

For details, please read more!

ImageNation - From scratch to a website
On Dec 15, 2013

We built this website for Database & information System class. It is an image sharing website, similar to Pinterest. We hosted our website on Amazon EC2, and our database on Amazon JDBC. Because of the credit issues, this website is no longer online. We used node.js to build our server, Oracle as our database, and jQuery for fron-end design. We wrote our own MVC framework and embedded REST in our website.

For details, please read more!

A Bag of Words
On Dec 6, 2013

How precise we can predict a rate of a restaurant only given all the words in that review? If we choose the correct method, the answer is it will be very accurate. This project is the final project of Machine Learning class, for which we need to train a prediction model on 20000 revealed data and make the prediction on 15000 unreleased data. We finally won the fifth place with a time consumption only one third of the top 3 teams.

For details, please read more!

A Lightening Fast Panorama
On Nov 17, 2013

6 653x368 images stitch togather in 12 seconds on a dual core i5 laptop, could be faster if Matlab can provide better image display function that can take less than 6 seconds to show stitched high resolution image. Meanwhile, the images are blended perfectly so that it is very hard to find the seams if there are any. I developed this mosaic algorithm for Computer Vision class project, and I am now discussing the possibility of a potential publication with the class instructor, Dr. Jianbo Shi.

For details, please read more!

Image Morphing
On Oct 15, 2013

It's a project for Computer Vision. I had a lot of fun doing this project. And I think we can use this method to design an app to predict what the child of two selected people looks like!

For details, please read more!

Amazing Circuit
On June 10, 2013

It is a PCB I designed for my senior design. The whole project is to redsign a solar cell powered mobile robot which can drive in unknown environment. I did lots of echanical design, PCB design and of course coding.

For details, please read more!

Cirular Logo Impainting
On April 25, 2013

This is the final project of Machine Perception class. The main focus of this project is feature (ellipse) detection in Hough space and homography transformation.

For details, please read more!

An old fashion stitching
On April 10, 2013

Stitching images in rectificated space, my first touch of OpenCV.

For details, please read more!

Canon on a 8-bit microcontroller
On Dec 20, 2012

Large range frequency generation from a small and not so power 8 bit microcontroller. So I decided to play some music with it. Hope you enjoy the music from pure sine waves!

For details, please read more!

Robockey2012 from MEAM Design on Vimeo.

Robockey Game 2012
On Dec 12, 2012

A famous competetion in engineering school at Penn is also the final project of Mechantronic class. Almost a week's hard work without sleeping, we designed three little beasts to fight in this epic game.

For details, please read more!

Puma Light Painting
On Oct, 2012

This is a project for Robotics class. We used old Puma robot to paint an image in the air. We choosed to paint a 2D image because it is easier but more interesting. We did the trajectory optimization of PUMA robot end effector so that we were able to get as smooth trajectory as possible.

For details, please read more!

Freescale Cup Intelligent Car Racing
On Sept 20, 2011

My team mate Chris (Chuang) Wang was putting the car on the track! We won the first prize in this competition, but more important, we learned a lot. I was in my sophomore year, and from then I never stopped creating and designing.

For details, please read more!


Name Email Message