My favorite course from the Spring of 2010 was called Learning in Robotics and was taught by professor Dan Lee. It was a rapid fire survey of topics using machine learning to improve common robotics algorithms. A simple example was training a color classifier to locate an orange ball in images. We trained our classifier on a number of input images and used that data to build a gaussian mixture of means model of the color range. Other projects included making extended and unscented kalman filters, using particle filters to perform slam, drawing paths through satelite images and combining these aspects in a project not unlike the DARPA MAGICC 2010 competition. My final project was a frontier exploration algorithm for a robot. It built an internal map of the environment and calcualted frontiers that were likely to produce more information to explore. The two videos below show the robot in the 3d simulator and a matching video that shows the internal map being produced in matlab.
In the Fall of 2009 I took a computer vision class with Jianbo Shi. We had projects on topics such as image stitchting, edge detection and object recognition. One of the videos I made for the class dealt with image morphing. Taking one image and turning it into another in a way that makes sense. You can see the short video below.
I stitched together a couple of pictures of Philadelphia's beautiful city hall using an algorithm I wrote as well.
One of my favorite classes at UPenn was mechatronics with Professor Fiene. He had a penchant for giving us fun projects with fun deadlines as well. We built light trackers using op-amps to do PID control, pwm robot game controllers using 555 timer chips to produce pwm signals, designed circuit boards in eagle and put together various other small robots. For the midterm project our professor had us build segway robots using a cheap (and very noisy) accelerometer. We blindly assumed that since the sensor had been assigned to us it must be up to the job of creating a balancing robot. Took us about a day to put the robot below together and about a week of tuning parameters to get to balance like it does in the video. Overall a fun learning experience in why a good sensor might be worth the money.
The final project for the class was robockey or as it sounds robot hockey. An IR emitting puck was placed on an air hockey sized table and 3 vs 3 robot teams were pitted against each other in a double elimination tournament. An overhead camera was used with fiducial markers to provide rough absolute locations of the robots on the field. Each robot's microcontroller also had the ability to communicate wirelessly with each other. Using a laser cutter my team built two different styles of robots and got to the semi-final round before losing to the winning team.
The summer of 2007 I built a 4 player arcade machine with my twin brother. It was designed in CAD to split into three pieces for portability. The design hides a computer in the base and has an LCD monitor as the screen. The computer runs a variety of pieces of software to emulate an arcade machine and to allow a person to select which game to play using a joystick. The arcade system hardware came from a supplier of arcade machine equipment and uses a special joystick to interface with a computer. The graphics were created in Photoshop and pieced together on 8.5 "x 11" pieces of photo paper. When it is necessary to use the computer portion a pull out drawer houses a keyboard and a mouse.
When the 2007 DARPA Urban Challenge was announced a group of undergraduate students from Princeton decided to enter a vehicle. Having entered in the 2005 Grand Challenge the team built upon the experience to put forth a competitive vehicle. Our entry was a 2005 Ford Escape Hybrid. We added stereo cameras from Point Grey to allow the car to detect obstacles and other vehicles. A monocular color camera also from Point Grey was used to detect lanes. A gimbaled Delphi radar was used to detect cars and provide long distance sensing. The car's trunk had 5 rackmount computers to run code implemented in the Microsoft Robotics Studio framework. I helped integrated the Delphi radar, mounted sensors and actuators, built a battery monitoring alarm circuit and programmed navigation algorithms.
After the success of the arcade machine my brother and I decided to build another entertainment device. This time I dreamed up a touch screen jukebox that would light up to the music. During the summer while working at inch-inc. I had created a portfolio of images to aid in our design efforts for a nostalgia inducing phot obooth. While making this portfolio I had grown fond of the art deco period and had decided to take inspiration from a mirror I saw. The jukebox houses a touch screen from a kiosk, a computer, integrates with a digital stereo system and has a circuit to turn the lights on and off to the music. It was constructed from wood, acrylic and polystyrene.
The summer of 2006 I was accepted into the NASA robotics academy at the Goddard Space Center. Through a partnership I worked with a team at Anthrotronix inc. a company which developed a number of interactive products both for children with disability and for soldiers who needed better control of their robots. My team further developed Cosmobot, a robot designed to help children with autism spectrum disorder learn social skills. The idea was that while a child with autism may struggle with social contact with people, the same child may connect very well with a computer. A robot can bridge this gap, by imitating some aspects of a person yet being a piece of technology that is easier for a child with ASD to connect with. To help Cosmobot achieve this goal I added a camera, modified his structure, upgraded his computer interface, programmed microcontrollers and developed a series of activities to interact with a child. At the end of the internship we tested these activities with a child who had ASD to great success.
For a class at Princeton myself and my team developed a product to be sold through As Seen on TV products. A partnership with Telebrands allowed us to have access to a team of people who had brought hundreds of products to the television market for 25 years. The product my team came up with was called the GO!Mug. It was a collapsible thermos that kept the aspects people liked about high end coffee tumblers but could shrink to fit into a large pocket or purse. We built it out of an innovative coil spring that made the mug feel like a solid metal thermos. After developing a working product we produced a commercial which was aired on TV to gauge commercial interest.
In response to a professors desire to develop products that could sustainably help developing nations myself and a group of students made a bicycle powered water pump. After researching water pumps we decided on a centrifugal pump design that could interface with any standard bicycle by replacing the rear tire with a belt. The belt would drive a shaft connected to the pump. The pump would be submerged in water and could under regular bicycling produce between 5 and 7 gallons of water per minute with 10 feet of head. This was similar to products on the market but could be produced for roughly 10 dollars.
During my time in high school my passion was FIRST robotics. Many of the skills I now use regularly I learned by manically assembling a robot each year in just 6 weeks. Skills as varied as procurement, CAD, machining, web design and team leadership all were required to complete a robot on time and to have it compete successfully. My team also made a strong effort to reach out and deliver FIRST's message to inspire students to become scientists and engineers. I still continue to mentor the team when I have time to pass on the knowledge I learned and give back to those helped me get where I am today.
