Andrew Trusty

My final project in Greg Wilson’s Topics in Software Engineering class this semester is a web application called Pubfeed.  Pubfeed is a tool that allows researchers in academia and industry to keep track of research in any areas that they are interested in.  All you have to do is tell it what research publications you like and it will generate a news feed of related research that is constantly updated.  All the feeds created on Pubfeed are public so you can view and subscribe to other peoples feeds.  You can monitor all your subscriptions via RSS or you can just check  the website every now and then.

The idea behind Pubfeed was Greg’s originally, I just took it and ran with it.  It is essentially a meta-search tool that just re-queries search engines periodically to check for new results based on user’s favorite publications.  The current implementation uses the DBLP and Google Scholar databases but I hope to add other data sources in the future.  In a sense, Pubfeed is actually quite dumb because all the heavy lifting in finding relevant publications is done by DBLP and Google Scholar and Pubfeed just aggregates their results with some basic filters.

There are already a number of interesting feeds on Pubfeed ranging from topics in computer science, medicine, music and economics.  Check out my subscriptions on Pubfeed and go create your own feeds!

I just got the news that the research paper I wrote from my senior research project was accepted at the Fourth Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE-08)! I guess this means I’m a real scientist now. Too bad I’ll be busy starting grad school in Toronto when the conference takes place at Stanford.

The work was a continuation of my involvement in the Cognitive Computing Lab (CCL) at Georgia Tech. Building on my previous experience with the and the CCL’s case based reasoning system, Darmok, and the Stratagus game engine, I developed an offline plan adaptation algorithm under the direction of Santi Ontañón and Ashwin Ram. If you’re interested you can read the full paper – Stochastic Plan Optimization in Real-Time Strategy Games.

I started working with a new research group in the Cognitive Computing Lab under Ashwin Ram this semester. The project I am working on is concentrated on using Case Based Reasoning techniques to easily develop AI opponents in video games. We are using Wargus, an open-source mod which allows you to play Warcraft 2 using the open-source Stratagus game engine, as the platform for our CBR research.

My contribution to the project was to develop a map classification system for Warcraft 2 maps which would provide additional features for the CBR engine. The system is a joint project between my Pattern Recognition class professor Jim Rehg and the CCL researchers Santi Ontañón and Manish Mehta. It was also a good starter project for getting more familiar with the architecture of their system since I plan on continuing to work with the group for my senior research project.

Continue reading…

I just finished a semester of research under Tucker Balch developing software to run on the Georgia Tech entry into the DARPA Urban Challenge. The goal of the DARPA Urban Challenge is to build an autonomous automobile which can navigate an urban environment as a human would. It was very interesting and challenging work with alot of other very motivated undergraduate and graduate students working to build out the platform to drive the Sting Racing team automobile. The students worked in a variety of teams concentrating on specific areas including Visual Odometry, Learning by Example, Pose Estimation, Health Monitoring, and Laser Scan Matching.

I worked in the Learn by Example team where we sought to use instance based learning to associate image-action pairs. Our team developed and tested a variety of different approaches for learning actions from images and efficiently matching to images in image databases. For my part, I implemented a smaller component of our teams overall architecture involving pre-processing of the images obtained from cameras mounted on the automobile.

I spent a lot of time researching different image classification techniques. At first, I was actually attempting to provide a very detailed classification of each image by labeling areas of the image as one of six classes: pavement, white lines, yellow lines, buildings, plants, sky. Given the real-time constraints and the high error of this approach I decided to modify my goal. After playing with a few other ideas and getting feedback from my teammates I decided to cut the number of classes down to two in order to effectively create an image mask identifying the road in the images. In this way my piece served to reduce the amount of data that needed to be processed by our other approaches.

My basic approach was inspired by a paper by Bischof, Schneider, and Pinz. The idea is to use a neural network to classify pixels of an image based only on the data available at each pixel. I used the Weka Machine Learning toolkit to train and test my neural networks. The image above shows the results of a six, three, and two class classification with each class represented by a color overlaying the original image. By the end of the semester I was able to develop a functional component that when given a camera image was able to output an accurate mask identifying the road in the image in real-time. I hope to be able to do future work on this component in order to enable it to learn online using the laser sensor data, parallelize the code to run on multi-processor systems, and output a confidence measure for the image masks.