Ultimate Auto-Pilot
by David Pescovitz
 Printer-friendly
version
UC
Berkeley
civil engineering professor Raja Sengupta
|
In the near
future, fleets of small airplanes may traverse our skies monitoring
traffic conditions, collecting data from environmental sensors,
and scoping out forest fires. The unusual thing about these
aircraft is that their cockpits will be empty. UC Berkeley
civil engineering professor Raja Sengupta is leading a College
of Engineering project to build intelligent guidance systems
for the next generation of unmanned air vehicles (UAVs). In
August, Berkeley technology enabled a UAV equipped with machine
vision to autonomously navigate a road for the very first time,
even in cloudy and rainy weather the researchers ran into at
a desert test site.
Sponsored by the Office of Naval Research's (ONR) Autonomous
Intelligent Network and Systems (AINS) program, the August demonstration
took place on a desert road outside of Tucson, Arizona. Amazingly,
Berkeley's effort to outfit UAVs from Advanced Ceramics
Research Corp. with machine vision began just four months before
the successful demonstration.
Professor Raja Sengupta describes Berkeley's UAV system to Congressman Kurt Weldon at the test site in Arizona. (courtesy the researchers)
|
Today's
UAV are just a few meters in length and can remain aloft for
several hours at a time on a small tank of diesel fuel.
The aircraft cost approximately $20 thousand each and are outfitted
with Global Positioning System (GPS) receivers that provide
location information. The user enters GPS waypoints that the
UAV uses
to get from point to point. According to Sengupta, the problem
with GPS is that it's just not smart enough for most
real-world applications. Take traffic monitoring, for example.
Earlier this year, Sengupta had been intrigued by demonstrations
where UAVs collect freeway traffic data much like manned news
helicopters do today. UAVs make sense because they keep people
safely on the ground and are less expensive to operate.
"I wondered why these are not commercial systems," says
Sengupta, the co-principal investigator of Berkeley's new
ONR Center for Collaborative Control of Unmanned Vehicles. "Then
I realized that it's very difficult for a UAV to follow
a road. GPS errors can cause a UAV to veer off its path quite
easily."
The Berkeley team's approach is to augment GPS with machine
vision software and a $120 off-the-shelf video camera. The
challenge, Sengupta explains, is for the computer to discern
the road
from the rest of the terrain from altitudes up to several hundred
feet.
Unprocessed view of the test road from the UAV camera. Inset shows the result after image processing to find the road and lane boundaries. (courtesy the researchers)
|
The solution is a two-step process devised by Zu Kim, a researcher
with the UC Berkeley-based PATH (Partners for Advanced Transit
and Highways) program. First, the software distinguishes the
road from the surrounding area based on differences in contrast.
In the desert, for instance, the asphalt is much darker than
the sand. Next, the lane boundaries are identified based on the
lightness of lane markings as compared to the asphalt. Once the
boundaries are located, the plane follows the lane from above.
"GPS will get the plane in the vicinity and once our system
locks onto the road, the plane can adjust itself regardless of
what
the GPS says," Sengupta explains.
Following the success outside Tucson, the Berkeley team is attacking
two more UAV navigation challenges. The first is obstacle avoidance,
historically a tough problem in computer vision. Indeed, another
Berkeley research project is focused on a vision system for trains
that detects any obstacle on the tracks, a car or pedestrian
for example, and alerts the engineer in time to avoid a collision.
Sengupta hopes to adapt the same technology, based on computing
a depth map of what the camera sees, to air vehicles so that
they may steer around them and then quickly return to their original
flight path.
The final prong in the Berkeley UAV research is focused on
what Sengupta calls the "canyon problem." While monitoring
traffic in an urban setting, a UAV may be forced to fly between
buildings lining the road it's tracking. One intriguing
solution was inspired by the navigation of bees. The insects
are able to fly down a corridor without bouncing between the
walls by determining if they seem to be moving past both walls
at the same rate. If there's a discrepancy between walls,
the bee adjusts its trajectory. Sengupta hopes that borrowing
this biological principle behind bee flight will lead to a computationally
quick way to deal with "urban canyons."
Solving all
three problems could open up an entire realm of UAV applications
far beyond traffic monitoring. Sengupta envisions
environmental scientists periodically deploying UAVs to wirelessly
gather data from remote environmental sensors--in forest canopies
or marine areas for example--that can only transmit short distances
due to limited battery power. UAVs could also safely and untiringly
patrol forest canopies, enabling early-stage fires to be quickly
detected and quenched before they blaze out of control.
"Our first test was a success because of the amazing students
from civil and mechanical engineering who worked in the desert
for three weeks in 120 degree heat getting the thing to fly," Sengupta
says. "Now we've reached a critical mass and the
ideas and applications are really bubbling up."
Raja
Sengupta's home page
"AINS
ain't toy airplanes" (Tucson Citizen, 8/21/03)
"Karl
Hedrick named PI of new ONR center"
Zu Kim's
home page
Partners for Advanced
Transit and Highways (PATH)
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.
Media contact: Teresa
Moore, Lab Notes editor, Director of Public Affairs
Writer, Researcher: David
Pescovitz
Web Manager: Michele
Foley
Subscribe or send comments to the Engineering Public Affairs
Office: lab-notes@coe.berkeley.edu.
© 2003 UC Regents.
Updated 9/29/03.
|
