
Jana Kosecka, associate
professor of computer science, shows off "Elvis" (right), the
department's largest robot.
|
|
Computer Science Research Focuses
on Robotics
By Robin
Herron
The movie A.I.
notwithstanding, don't expect to see human-like robots in your neighborhood
anytime soon. "We're a very long way from that--two hundred years or maybe
more," says Sean Luke, associate professor of computer science.
But building a robot
like David in A.I. isn't Luke's goal. He and other professors in George
Mason's Computer Science Department are investigating various facets of robotics
that may have more immediate applications.
"Up until three or four
years ago it was prohibitively expensive for us to get involved in robotics.
Now we have robots that are robust and adaptive--and affordable," says Kenneth
De Jong, professor of computer science.
Computer Science faculty
now use off-the-shelf robots designed for research, which range from small
$200-$300 robots built from Lego kits to large barrel-size robots that cost
almost as much as a Volkswagen Beetle. Most of the research, however, is
done by simulation, with an actual robot used only for final testing, De
Jong says.
Luke, whose interest
is in evolutionary computation, has used only his computer for some robotics
research. Using the model of a soccer game for his study, he created white
dots--softbots--on his computer screen. These softbots became the soccer
teams. The computer developed programs and then played the teams against
each other, selecting the better ones and "breeding" them, thus creating
new generations of improved programs. In the early rounds, the softbots all
went to the ball. Eventually, they "learned" that if one or two stayed back,
the ball would eventually pop out of the pack and they could block it from
going into the goal. Finally, the computer teams learned that their chances
to score increased when "players" spread out into strategic positions.
The learning process
that Luke demonstrated showed that robots could be programmed to work in
teams with other robots. The process' practical applications might be used
in a battle simulator or search and rescue efforts. Another application might
be in building space stations, where sending each command to a robot from
earth is tedious and time-consuming. "You need to have the robots be partially
autonomous so they can coordinate with humans and other robots," says Luke.
Could a robot be programmed
to assist an elderly person in a home setting? To do that, it would need
a kind of computer vision that would let it map out three-dimensional objects
so it could navigate the home. Using the department's largest robot, dubbed
"Elvis," Jana Kosecka, associate professor of computer science, is developing
algorithms to help the robot build a geometry of its world.
Elvis is equipped with
ultrasound sensors to "see" objects at long distances, infrared sensors to
sense objects up to a few feet away, and a video camera to record what it's
"seeing." Elvis's short-term goal is to tour the Computer Science Department
without bumping into things. The long-term goal is for Elvis to interact
with objects. "We want to push the complexity of the kind of environment
they're in as well as the kind of interactions," Kosecka says.
Zoran Duric, assistant
professor of computer science, works on tracking human movement. Used with
a video camera, his computer programs recognize human behavior and human
gestures using color and objects' edges or boundaries, differentiating the
human subject from its background and tracking his or her movement. The programs
can also determine whether the subject picks something up or drops something.
Although it may sound
simple, tracking movement is actually quite complex, Duric explains, and
he compares it to developing notation for music or choreography for ballet.
Duric's programs could be used by a robot outfitted with a camera for surveillance
or safety purposes, such as monitoring children near a busy intersection.
The research could also be used in video games and animation, in searching
video to find specific events, or in sports analysis, for example, to help
a golfer analyze his or her swing.
Enthusiastic about the
possibilities of robotics, De Jong relates a list of practical applications.
"Robots have a long association with industry and the military," he says.
"Robotic vehicles can be used in dangerous or unpleasant situations, like
a nuclear waste site or the sewer system in New York City. A robot could
help long-distance truckers navigate or provide an alert mechanism because
a robot doesn't get sleepy or distracted.
"Almost every object
has a computer chip and can be programmed to be more intelligent or autonomous,"
De Jong says.
|