| Eye in the Sky
by David Pescovitz
A model helicopter is gently hovering in the sky above a field at
the UC Berkeley Richmond Field Station. Slowly, it begins its decent
onto a landing platform secured to a trailer. The landing is precarious
like the pilot is new to the hobby but ultimately
successful. Onlookers cheer but the pilot is nowhere to be found.
That's because the pilot is a computer.
Movie: Pursuit-Evasion Game with both aerial and ground
Movie courtesy Berkeley Aerobot
The Berkeley Aerobot (BEAR) is an autonomous aerial robot, designed
to fly and navigate and recognize and locate target objects on its
"You could imagine hundreds of these being deployed to fight
fires or conduct search-and-rescue machines," says T. John
Koo, visiting faculty at Berkeley's Department of Electrical Engineering
and Computer Sciences (EECS) and a principal investigator of the
Aerobot project. "With that many helicopters flying, the number
of people involved in operating them must be reduced. Basic robotic
functionality enables humans to worry about more high-level tasks
like where the helicopter should go and what it should look for."
The Aerobot first took flight in 1998. Koo, then a Berkeley Engineering
graduate student, spearheaded the University's first entry into
the Aerial Robotics Competition with support from EECS department
chair Shankar Sastry, still the faculty leader on the Aerobot project.
John Koo in his laboratory with a prototype Aerobot. (Click
for larger image.)
David Pescovitz photo
Today's Aerobots based on the airframes of remote-controlled
crop-dusting or hobbyist helicopters are a tour-de-force
of computer vision, embedded software, and control systems. Employing
Global Positioning Satellite technology that provides location information
down to two centimeters, the Aerobot steers itself to any pre-determined
waypoint. Then, advanced computer vision technology developed at
Berkeley kicks in to guide the helicopter safely back to Terra Firma.
But landing on the ground looks like childsplay compared to one
of the other research objectives. Funded in part a grant from the
Defense Advanced Research Projects Agency's Software-Enabled Control
program, the Aerobot is expected to be agile enough to land on a
moving aircraft carrier. As a testbed, the Berkeley team designed
a mechanical landing platform that simulates the rock and sway of
a ship on the open sea.
Indeed, simulation plays a key role in the Aerobot research. The
majority of Aerobot experimentation takes place indoors on computer
screens. One PC runs the helicopter's software code while another
emulates the dynamics of the helicopter and its onboard GPS and
intertial/navigational sensors. Other variables a digital
representation of wind, for instance add to the model's realism.
computer simulation with hardware in the loop, we can test the entire
system without risking the expensive experimental helicopters,"
says Koo, adding that a fully-outfitted prototype Aerobot can cost
in the tens of thousands of dollars.
The latest addition to the Aerobot's test regimen are Pursuit/Evasion
Games. The games consist of an Aerobot collaborating with several
ground-based mobile robots to chase a "renegade" mobile
robot. Whoever locates the evader first guides the others to the
location not unlike a police chase on the Los Angeles freeway,
albeit much slower. The Aerobot team's future computer vision research
goals focus on combining the GPS-based control system with 360 degree
videocamera technology for path planning and aerial obstacle avoidance.
Koo believes the Aerobot is ideal for locating objects other than
errant vehicles though. With the intense accuracy of their onboard
GPS system and the integration of additional sensors, Koo says the
Aerobots could fly inches over terrain and pinpoint the location
of undetonated mines. Another application Koo envisions is robotic
firefighting. For instance, a continuous swarm of Aerobots could
dump water on a forest fire without the risk of collision in dense
smoke, he says.
While real-world Aerobot applications are still several years away,
Koo is looking at other potential benefits of his embedded software
and control systems research. All aircraft, he explains, could benefit
if the numerous automated tasks such as the angle and speed
changes that must be taken into consideration when landing
were better orchestrated to achieve a high-level goal. To that end,
the Aerobot group is collaborating with industry partners like Boeing,
Northrop-Grunman, Honeywell, and Rockwell "to make aircraft
safer and more efficient."
"Sensors and actuators are the interfaces between computers
and the real world," Koo says. "But we need those two
worlds to be tightly coupled."
T. John Koo's
Lab Notes is published online by the Public Affairs Office of the UC Berkeley College of Engineering. The Lab Notes mission is to illuminate groundbreaking
research underway today at the College of Engineering that will dramatically change our lives tomorrow.
Editor, Director of Public Affairs: Teresa Moore
Writer, Researcher: David Pescovitz
Designer: Robyn Altman
Subscribe or send comments to the Engineering Public Affairs Office: email@example.com.
© 2002 UC Regents.