Robot Kits
Stiquito Kit
BattleKits
Robot Toys
Solar Kits
Robot Arms
Robosapien
Basic Stamp Kits
Lego MindStorms
Beginners
Books
Hobby Robots
Robot Sports
Electronics
Mechanics
Robot Minds
Books for Kids
Robots at Work
Microcontrollers
Advanced Books
Used Books
Real Robots
Robot Motors
Remote Controls
Robot Parts
Robot Tools
Magazines
Robot Videos
Robot News
RoboLinks
Contact
|
Source: Stanford University
It's a robot version of spy versus spy.
Stanford computer scientists have equipped "observer" robots with video cameras
and successfully programmed them to track "target" robots. The researchers now
are working on the more difficult problem of programming their observers to stalk targets
that are attempting to evade pursuit.
Naturally, the Army is interested, and is providing financial support for the project.
Observer robots that can automatically track potentially hostile targets could be real
lifesavers for soldiers fighting in conditions of limited visibility, such as urban
environments.
Autonomous observers (AOs) as computer science Professor Jean-Claude Latombe and his
students call their electromechanical creations have other potential uses. In operating
rooms they could automatically keep video displays which surgeons depend upon for delicate
operations zeroed in on key tissues despite the obstructions created by the movements
of people and machinery. AOs could perform search and rescue operations in potentially
hostile environments; monitor operations in remote assembly plants; and supervise
automated construction efforts in outer space.
"Originally, we called them intelligent observers, but they aren't really that
intelligent," says Hector H. Gonzalez-Banos, a doctoral student in electrical
engineering who is working on the project. "So we named them autonomous observers
instead."
Computer science Professor Jean-Claude Latombe, left, and graduate student Hector
Gonzalez-Banos are framed by an autonomous observer on the right and a target robot on the
left.
The autonomous observer may not be all that intelligent, but it does more than simply
follow its target around at a constant distance. It must constantly calculate the
positions that it needs to assume to ensure that the target doesn't disappear behind a
column or down a hallway.
This kind of robotic motion planning is anything but simple. The robot first must be
familiar with the area in which it is operating, so Latombe's group gave its observer
robot the ability to map a new area when it first enters. The robot moves about
continually measuring the distances to walls and furniture with a horizontal laser range
sensor and it uses this information to create a two-dimensional floor plan. Next, the
robot uses a horizontal video camera to create a series of overlapping three-dimensional
views of the space. Finally, it combines these into a 3-D rendering of the space.
The observer robot carries a second camera that is focused on the ceiling. This helps it
keep track of its position by recognizing block patterns that the researchers have
attached to the ceiling in a grid pattern. "We could program the robot to recognize
the tile pattern, but this is a lot easier," says Gonzalez-Banos apologetically.
The target robot also sports its own black-and-white pattern stenciled on every side. The
observer robot uses this to identify it. In an associated project with Professor Ruzena
Bajcsy's group at the University of Pennsylvania, Latombe and his students are working on
a technique that will allow the robot to identify and track unmarked robots and people.
Human operators have a choice of two views on their computer screens. In one, they can see
the view from the AO's horizontal camera. But when the robot is zipping hither and yon and
panning quickly back and forth, this viewpoint can leave people pretty woozy,
Gonzalez-Banos says. So, in most cases, the operators choose the 3-D view that shows the
position of the observer and target robots within the representation of the space that the
AO initially created.
So far, the researchers have been experimenting with a single observer and a single
target. The robots, which were built by Nomadic Technologies of Mountain View, Calif., are
about 4 feet tall and resemble an upright tank vacuum cleaner without the hose. Shortly, a
new grant from the Army will give the researchers four additional robots. The new, smaller
robots will allow the researchers to devise methods for deploying multiple observers.
Latombe's group also is working with Steven La Valle of Iowa State University on the
problem of how to deal with a target that is trying to escape from surveillance.
One of the group's goals is to program groups of autonomous observers to work together to
locate designated targets. This would give AOs the ability to "sweep" a given
area for a target, robotic or otherwise, that is trying to avoid detection.
Given the broad range of possible applications for autonomous observers, the idea that led
to their development was surprisingly modest. "We were just looking for a way to
share robots with researchers in other laboratories," says Latombe. The cost of
building and maintaining research robots is extremely high, so finding ways to spread that
cost was quite appealing.
The Internet gives researchers an inexpensive way to program and operate robots from afar,
but it didn't have a good way to view exactly what the remote robot is doing when the
researchers are running it. The autonomous observer was Latombe's solution. By using one
of their robots to keep a second, target robot continuously in view and then feeding this
view back over the Internet, the long-distance researchers can observe what the target
robot does as it follows the instructions that they send via the Internet.
At the Monterrey Institute of Technology (ITESM) in Mexico, researchers headed by
Professor Jose Luis Gordillo are collaborating with Latombe's group on this project. The
Mexican scientists have programmed one of their robots to identify empty Coke cans sitting
on tables and sweep them into a waste basket that it carries.
To test the value of the autonomous observer, Latombe and his students modified one of
their mobile robots so that it can perform the same task. The Monterrey researchers then
wrote a program for the Stanford robot that allows it, at their command from Mexico, to
collect Coke cans placed around the Stanford lab. As the target robot wanders about the
Stanford lab searching for empty Coke cans (it only collects Diet Coke empties because it
can only recognize cans with white labels), the observer robot zips around keeping it in
sight. That allows the Monterrey researchers to see just how well their program is
working.
|
|