Designing - Rohit Unnam - Carnegie Mellon

So, when we were testing our code with all of the sensors and curcuit boards attached to the drone, we noticed a very troubling problem. Not only was the drone too heavy, it was just too messy. There were wires sticking out that frequently got caught in the propellers, random items attached to the body to help balance the weight of the drone and it frankly just looked really bad. So, I took it upon myself to design our own custom frame that would better facilitate our needs. This is the frame that they are currently using at the lab. Below is a picture of it.

There are 4 slots on the front, sides and back of the circle for the different sensors that we are using and slots to mount the circuit board and custom blade guard we have. In the end, we dropped about 70g of weight which increased flight time from 7 minutes to 15 minutes. It also helped keep the wires much more ruley.

Lesson in practical engineering - Rohit Unnam - Carnegie Mellon

So, Mr. Butzke, our project leader, took us out to teach us about 3D modeling, 3D printing and laser cutting. He showed us the laser cutter they had on campus as well as the 3D printer and the entire array of machines they had on campus for students to use. He gave us a tutorial on using CAD/3D modeling software which sparked my interest in the next portion of our project. Unfortunately, my phone was not working that day so I do not have any pictures for you.

Coding - Rohit Unnam - Carnegie Mellon

My first part of the project was coding. We needed a flight controller that would translate code commands into commands that the AR.drone would understand. We did this by utilizing the ARdrone_autonomy open source packages which did the translation part. All we needed to do was make a custom plugin that could understand the code that we were trying to use. It sounds a lot easier than it actually was. It took us about 2-3 weeks to finish this portion of the code. After we finished that part, things got a lot more slow. There wasn't much to do and the World Cup was airing so no one was really working, as you can see in the picture of us watching the USA vs Germany game.

After the World Cup was over we finished the collision_avoidance package and then began our testing, during which I found some major problems with the design of the drone...

Starting at the lab - Rohit Unnam - Carnegie Mellon

The first few weeks at the lab were mostly just orientation. I was introduced to the people in the lab and got to hear about all the cool stuff they were working on. Some were working on programming new algorithms to more efficiently path plan for various tasks, some were working the the PR2 robot and applying their planning algorithms to different tasks, and so many more that I can't list. My group however, was working with the commercial AR.drone. The 4 of us were the only ones doing that so we were pretty much our own sub-lab. Below is a picture of the other two interns in our group. We I asked about how to program the PR2 and they demonstrated by trying to get the $400,000 robot to shoot a nerf gun...

On the left is Ellis and on the right is Sam. They were pretty cool and knew a lot about programming and taught me a lot.

Rohit Unnam - Search Based Planning Lab - Carnegie Mellon - Lab Summary - Week 1

The first few days at the lab were somewhat slow as my team leader, Jonathon Butzke wasn't going to be at the lab until the second week. But, I got to meet all the people at my lab and see what kind of stuff they were working on. There were people working on a whole variety of things from getting a robot to maneuver a picture frame through a picture frame, to getting a robot to deliver drinks to desks in the lab. The people in the lab are extremely experienced. Most of them are Grad students, some graduated some interns from other colleges. My team head was a former air force pilot. There was even someone who went to the same middle school as me (Princeton Day School).

Here's a picture of the group:

]

There are about 10 others who don't work directly in the lab but are a part of (including the PI himself).

Chris Oh Week 6 - Gabrieli Lab, MIT

    On Monday of my final week, I had a meeting with Zhenghan to discuss my last five days at the lab.  She had suggested that I look at the ANOVA results and the OLS Regression results to look for any statistical significance or patterns.  She also recommended that I compile all the plots and regression results into one sheet on notebook viewer, which allows me to publish all my work this summer on a http address.  Because she was going to Netherlands that weekend for a conference, she wanted me to just work on my poster after I finished all of these things.

   After plotting all the language scores with the average durations and the five syllable durations, I took the OLS Regression for each plot to look at the p-values.  None of the 34 graphs had p-values lower than the confidence level (p=.05), but one came very close to it.  The p-value from the plotting of five syllable durations and the test scores from a test called CTOPP-2 Blending Words is 0.088, which was just slightly above the confidence level.  The Blending Words task measures the ability to synthesize sounds to form words, which similar to the Nonword Repetition task.  The plot showed that the subjects with higher score in CTOPP-2 Blending Words task took lesser time responding to 5 syllable words.  There was no correlation with the average durations, suggesting that the difference in the ability to formulate words showed up only when longer words were presented.  The OLS Regression also revealed that the coefficient of determination is 0.182, which suggests approximately 42.7% predictability.  

   The ANOVA results for the average durations for each syllable showed no signficance with the p-value of 0.456, which exceeds the confidence level by much.  

Average durations for each syllable:

5 Syllable Durations and CTOPP-2 Blending Words scores:

For all the graphs and regression data, visit: http://nbviewer.ipython.org/gist/chris0705/5f521a3ac296343a9595