 |
|
 |
|
Carnegie-Mellon's "Sandstorm" leaves
the starting chute at dawn |
|
 |
|
 |
|
|
|
|
The Golem Group's white pickup speeds up on a
straightaway |
|
|
|
|
|
|
|
|
Team DAD's "Dad Are We There Yet?" |
|
|
|
|
|
|
|
|
The 16-ton "TerraMax," sponsored by
the Oshkosh Truck Company |
|
|
|
|
|
Continued
from page
1
DECIDE
All of the sensor systems can produce "false
positives", seeing objects that are not really there.
There are even "negative obstacles," like the canyon
in Thelma and Louise! Figuring out which sensor to
believe over another, how to integrate all of the inputs to
produce a map of the road ahead in memory, and then how to
choose the right path to the next waypoint, those are the
real problems. The control software at this point is so complicated
that only practice can tell if the right choices are being
made by the bot. There are variables in the code that can
be adjusted to influence those choices, but the bots had an
amazing capacity to surprise their creators with unplanned
behavior -- especially the losers.
ACT
This is where the rubber meets the road.
And where the servo meets the steering wheel! Some teams,
like Team DAD, had it down so smooth, you'd swear someone
was actually behind the wheel. Other teams, like the Golem
Group, had lightning reactions, charging up to obstacles and
swerving at the absolute last moment. Still others kind of
meandered around. All teams had to be sure not to get into
an overcorrection loop, where each action requires an opposite
reaction. That's when a bot can shake itself to death or bounce
like a pinball from guardrail to guardrail.
Many teams could not reverse. TerraMax
sure could! One of the reasons it took 12+ hours to complete
the course was that it in spite of its steerable rear axles,
it often made K-turns around sharp corners, backing up many
times for one turn. Of course what was impressive was that
it knew when to turn like that and successfully did so for
TWO DAYS.
Up until the National Qualifying Event held
a week before the final race, Carnegie-Mellon was the
team to beat. Stanford handily demonstrated its winning strategy
at the NQE, and it was then that it was evident there would
be a real horse race in the Mojave. Also at the NQE, after
making it past many narrow obstacles with (relative) grace
and surprising agility (that 16-ton monster can stop on a
dime!) no one doubted that TerraMax would (eventually) finish.
The real dark horse was Team Gray.
The vehicle for Team Gray was a Ford Escape
Hybrid. It's most interesting and visible feature was
a huge table of solar panels mounted on the top. At about
300W of capacity, they were used as the only way to charge
the 24V battery system that powered the LADAR. Computing power
came from a main microprocessor that was shock-mounted and
designed to run on 12V. The team soon learned that the hybrid
drivetrain has almost no starting torque. During trials, the
Escape could not pull its chase vehicle out of the mud, but
the team hauling on a rope did. The Escape also once came
to a stop in sand, piled up a couple of inches in front of
each wheel, and could not get out. Couldn't even spin the
wheels!
Nonetheless, the Escape had enough of the
right stuff to finish the race. (This may say more about
the course than the vehicle.) In many ways, the Gray's strategy
represented a minimum approach (and certainly the smallest
budget). Only the right technology and software was brought
to bear on the problem. Team Gray just might have a thing
or two to show the others.
Will there be a race next year? It's
unlikely. DARPA got what it wanted, the technology to build
autonomous vehicles for use in combat situations. Considering
the difficulty of the course, these bots are ready for duty
in Iraq, but not a place like Afghanistan. Maybe next year,
someone will organize an Extreme Challenge…
Mark
Helmlinger
October 16, 2005
