Report From the Lab: Julia Ferraioli '07 and Leslie McTavish '08 Teach Robots How to Learn

This summer, Julia Ferraioli ’07 and Katharine E. McBride scholar Leslie McTavish ’08 worked as research assistants in Associate Professor of Computer Science Douglas Blank's developmental-robotics laboratory along with Swarthmore student George Dahl ’08. Ferraioli and McTavish are both majoring in computer science, but their only previous experience with robotics and artificial intelligence was outside the classroom; they had been part of the team that created "Bubbles," a robotic blimp that made a strong showing in a contest at Drexel University last spring. It was an exciting summer in Bryn Mawr's robotics lab: In July, Microsoft announced that Bryn Mawr had been chosen along with the Georgia Institute of Technology to host a new Institute for Personal Robots in Education. Just a few days later, the students accompanied Blank to the American Association for Artificial Intelligence's 21st National Conference on Artificial Intelligence, where they presented their research findings at a poster session. They also demonstrated Bubbles, the robotic blimp, and won a prize for innovative hardware design. Ferraioli and McTavish's account of their research activities follows.

Blank, Dahl, Ferraioli and McTavish in the lab with Aibo and Bubbles              (photo by J. Fei '08)

The first day of summer vacation began with a bit of apprehension for both of us. We had jumped straight from finals into our summer research project in developmental robotics. Neither of us had taken a course in artificial intelligence or worked in depth with Python, the programming language in which Pyro, a robotics research tool, is written. In the past semester, we did gain some experience with robots by helping build and program a robotic blimp, but for our summer research we would be using either simulated robots or Sony Aibos (Aibo is a robotic dog. "Aibo" comes from artificial intelligence robot, and the name means friend in Japanese). New experiences, while they can be exciting and challenging, can also carry an element of anxiety.

Before we could think about developing a project plan, we had to start by learning how our subject has been approached by others. Douglas Blank, our adviser, suggested articles to read, primarily on developmental robotics. Developmental robotics is a methodology in which a robot starts out with only a primitive seed program and eventually learns sophisticated behaviors on its own, without explicit programming or memorization. At first we were discouraged by the sheer number of papers available, since the initial articles discussed concepts which then had to be researched. After all, knowing what you can do with a neural network is not much use unless you know the ins and outs of one. One of the most important lessons that we took away from our initial research is that you should probably allow yourself at least as much time to read and comprehend an article as it took for the author to write it.

As we each found areas of research that were interesting to us, our paths diverged. No team member was working on the same task, which made our weekly meetings very interesting as we all presented something that was new to everyone else. Toward the end of the summer, we combined some of the concepts we had worked on individually to reproduce an experiment in which the Aibo tries to learn to associate its motor values — that is, the positions of its various body parts — with its sensory readings of the environment. This experiment was one of the posters we exhibited at AAAI conference in Boston.

At the competition where we had entered our robotic blimp, Dr. Oh from Drexel invited us to showcase it at the AAAI (American Association for Artificial Intelligence) Conference. Dr. Blank thought that we could participate in the scavenger hunt and present our Aibo experiment as well. At the beginning of the summer, still overwhelmed and unsure of our grasp of key concepts, we were not sure that we would be able to adequately explain what we were researching. However, when it came time to present our posters, we felt confident in our skills and everyone walked away with all of their questions answered. The scavenger hunt seemed to be more of a showcase than an actual competition, due to the fact that few of the robots actually performed the required tasks. We did, however, win an award for "Innovative Hardware Design" for our blimp, Bubbles.

The conference was exhilarating and exhausting at the same time. We met so many people from different institutions (academic and corporate) around the world, and it felt gratifying that professors came to us for explanations of our posters. We did not get to attend too many of the talks, because we felt obligated to answer the many (often constant) inquiries about our research. Whether the interest was due to the fact that a huge purple-and-blue blimp was hovering near our table or that two women were answering most of the questions was not quite clear. However, the constant questions not only clarified issues for the inquirer but helped us to improve our discussion techniques. We learned just what aspects of our projects that people found interesting. After just the first day, we were able to refine our explanations to concise yet informative descriptions of our work.

The research that we did this summer gave us great insight into the world of developmental robotics — its challenges, appeals and limitations. AAAI was an invaluable introduction into how research is presented in both an academic and corporate setting, as well as hands-on experience in presenting research to those familiar with the field yet not with the specifics of our particular area. While our summer experience had its challenges, we ended it with confidence in our research and the knowledge we gained from it.

<<Back to Bryn Mawr Now 9/14/2006

Next story>>