|
Teaching Students to Teach Machines
By Karen Young Kreeger
Robotics and the liberal arts may seem like an oxymoron. But two members of the computer science faculty at Bryn Mawr are leading a quiet revolution of sorts to change that. In collaboration with colleagues at other institutions, they have developed a series of innovative artificial intelligence (AI) courses involving hands-on experiments with robots. This year, Associate Professor Deepak Kumar, Assistant Professor Doug Blank and Swarthmore College colleague Lisa Meeden received a $400,000 grant from the National Science Foundation to develop a cutting-edge robotics course that would incorporate new ideas on how to teach machines to learn.
The three-year project, which includes hardware, software and curricular components, is an extension of a robotics course that Kumar and Meeden introduced in 1996. Using LEGO toys and a Handyboard (a small controller board developed at MIT), they developed course materials that enable AI professors to use affordable robotics in the classroom and give students practical experience in writing algorithms and programs that make robots do different tasks.
“Today, if you go to FAO Schwartz, you can buy the LEGO Mindstorms robot kit for less than $200,” says Blank. “My 7-year-old son has a kit like this. Even he can do interesting things to examine robot control.” Of course, the programs used in AI classes are more sophisticated than those used by Blank’s son, and have proved to be an effective means to introduce robotics to undergraduates.
The Next Gen in Robotics Education
In the past seven years, more than 100 universities worldwide have adopted the LEGO/Handyboard-based course materials in their classrooms. But Kumar, Blank and Meeden wanted to introduce advanced concepts in cognitive science into their courses — for example, to explore neural networks, which are complex programs patterned after the neuronal connections in mammalian brains. “We realized that there was a hardware limitation to the LEGO-based robots because their sensors are very simple,” says Blank. “We wanted to explore robots with cameras and sonar sensors.”
Katherine E. Kline ’02, a graduate student in systems science and mathematics at Washington University, St. Louis, took Blank’s robotics course (Androids: Design & Practice) in her senior year. “It’s interesting to teach a machine, which is inherently stupid,” says Kline. “I have always been curious about how we think, how we learn. To answer those questions, some people do psych while others do biology.” Kline opted to explore human cognition and learning through robots.
“There was such a wide variety of people in Doug’s class. Their projects made me think about the subject in whole new ways, and that helped me decide to pursue these questions at a graduate level,” Kline says. Her class project dealt with topological mapping. Kline and her project partner wrote a program that would make their robot maneuver in a maze and avoid getting stuck in corners.
sensors might not always work, or you might be working with a reflective surface that throws your measurements off,” she explains. Kline is already excited about the prospects of working in robotics after grad school, and hopes to blend her interest in medicine with robotics.
Beyond LEGO
“With the help of the NSF grant, we hope to bring research-level robot experiences to our students,” says Kumar. “We’re going to continue using the LEGO model in our basic AI courses, but we’re creating more advanced courses with sophisticated robots as a follow-up.”
What makes the new courses work is Python, a programming language that can be used on different types of robots — a universal translator, as it were, which makes it easier for novices to incorporate advanced neural-net types of programs. Using Python, Blank and his colleagues have developed a new robot programming environment called Pyro. Before Pyro was available, researchers would have to write unique code for each type of robot they worked with, which for the most part was only applicable to their particular research interests.
A new generation of robots is also being used in the advanced courses. The smallest is essentially a hockey puck on wheels, petite enough to run experiments on a desktop. “These robots, named Kheperas after the Egyptian scarab, are ideally suited for experiments that use programs that take a long time to run. We don't have to worry about a robot getting stuck in a corner or falling down the stairs in the hallways,” says Kumar. “We can leave it running on our desk for hours and it’s safe.”
Another new robot is about the size of the Mars Pathfinder, approximately two feet long, one foot high, and one-and-a-half feet wide, weighing in at 90 pounds. The largest model is the size of a trash can and weighs 250 pounds. All of these robots are equipped with attachable cameras so students can do computer vision experiments.
This semester, three new courses that use the Khepera robots are being offered — one each at Bryn Mawr, Swarthmore and the University of Massachusetts, Lowell. In these courses, students are learning how to write programs that enable the robots to learn about their environment.
“Having a robot do exactly what it was programmed to do is fun. To have it autonomously learn to do those tasks is even more amazing,” says Kumar.
About the Author
Karen Young Kreeger is a science journalist who writes on biomedical and women’s health topics, as well as careers in science. Her most recent work has appeared in Bioscience, Genome Technology, Muse and The Scientist.
| 
