The Golem Group's DARPA Vehicles: from left,
Golem 1 and Golem 2.
Vehicles That Drive Themselves? UCLA Engineering
Pushes Autonomous Control to the Limit
2005 DARPA Grand Challenge
Contenders Get Their Start on TV's “Jeopardy”
What is the DARPA Grand Challenge? A
group of vehicles traveling under their own steam crossed the
desert on Saturday, October 8, in a high-stakes government race
worth $2 million. Among them was the team from the UCLA Henry
Samueli School of Engineering and Applied Science – a talented
group that started out with a desire to win and a jackpot garnered
by one of its members on the hit TV game show “Jeopardy.”
Sponsored by the Defense Advanced Research Projects
Agency, or DARPA, the Grand Challenge was conceived last year
to encourage the development of unmanned vehicles for combat.
Held in the desert Southwest, the teams must develop an autonomous
ground vehicle to finish a designated route the quickest within
10 hours to win. The initial prize money, doubled after none of
the 15 contestants in last year’s race were able to complete
the difficult course, was a nice incentive, but for UCLA it was
not the central motivation.
“It’s really about the challenge of
building a machine that can essentially drive on its own more
than it is about the money,” says Jason Meltzer, one of
the returning members of the group this year whose college quiz
bowl colleague Richard Mason sank his $30,000 Jeopardy winnings
into outfitting the group’s first car. “The race is
a strategy game. It takes knowledge, a lot of preparation, timing
and luck. In this case, it was a team effort.”
This year’s DARPA team, which was a consortium
of The Golem Group – the company Mason co-founded with teammate
Jim Radford – and UCLA, included faculty and graduates from
computer science, mechanical engineering and electrical engineering,
among them UCLA Vision Lab director Stefano Soatto and mechanical
engineering professor Emilio Frazzoli.
Eagle Jones, another returning graduate student
from last year who spent considerable time and effort on the project,
says the group was well positioned for the second race, particularly
because they come from diverse engineering, scientific, and cultural
backgrounds, all of which added new perspectives on problems that
a more concentrated team might miss.
“The hardware and integration was well planned
and executed. Our software was been developed far beyond last
year. A large number of sensors are integrated into the vehicle,
including a vision system from MobilEye that finds off-road paths,”
Jones explained.
The team was hoping that all of these improvements
would spell success. Last year’s unmanned vehicle traveled
5.2 miles, and was one of the best performers in the field, despite
a fatal problem with a throttle replaced just before the race.
For their second attempt, the team designed and
equipped two vehicles simultaneously – one as a contender
and one as a back up. The vehicles cannot be controlled remotely
during the race and must rely on global positioning, various sensors,
lasers, radar and vision systems, or cameras, to orient themselves
and detect and avoid randomly placed obstacles.
At the beginning of this year's race, Golem 2
raced out of the start chute in the seventh pole position, and
all seemed well. Tragedy struck at mile 34. The team had overallocated
the computer's memory, but this did not become apparent until
much of it already had been used to record data. The computer's
memory filled up and the program simply quit.
The team says that because of their high speed,
UCLA was on pace to finish the race significantly ahead of Stanford
and the other teams who completed the entire race. The team's
website, updated after the race results naming Stanford the winner
were made public, is philisophical - "It's entirely possible
that we would have won the two million dollars with a tiny change
in our memory allocation. But that's the way it goes."
Nearly 200 participants from across the US and
Canada and several foreign countries applied for the Grand Challenge
this year, a much higher number than previously. The contenders
were whittled down in various challenge races to the final group.
For more information on UCLA’s
Golem Group, visit http://www.golemgroup.com,
or to find further information about UCLA’s vision lab,
go to http://www.vision.cs.ucla.edu/.
For more information on the DARPA
Grand Challenge, visit http://www.darpa.mil/grandchallenge/.
###
10.11.05
-Melissa Abraham
|
