A wonderful example of why "autonomous" vehicles will *not* be
practical on our battlefield for many years to come. Clearly, he has
no problem tracking objects against a blank wall. But, I'll guarantee
that the same experiment conducted against an open background would
result in failure. Motion detection and tracking outdoors is an
extremely difficult problem and I personally know one video security
company (www.sightlogix.com) that has spent many millions of dollars
to accomplish that goal. Their system uses a half-dozen high-speed
processors to allow them to filter out unwanted inputs, like swaying
trees, leaves blowing across the lawn, birds, etc. to allow the camera
to focus on targets of interest.
But, for argument sake, let's assume that the system in the video
works perfectly on the battlefield, tracking everything that moves.
Then, the system has to solve a classic range and trajectory problem
just like the earliest fire control computers had to solve. Range
resolution can be done with a simple sensor (e.g., sonar) over short
distance, but at battlefield distances (up to 75 feet) it's a much
more difficult problem. Throw in the fact that our tanks are very
close to the ground, with a maximum height only 4 times the grass
around them, and you'll need a very good noise filter as well. Once
you get the proper range, solving the trajectory problem is a
relatively simple chore, but without any fire-control feedback you'll
have to rely solely on the accuracy of sensor information to actually
hit the target.
But, for argument sake, let's assume that even those problems can be
solved. The system now has to decide "which" target to shoot at ...
discerning the importance of each target (clearly, a tank bearing down
on you is more important than a supply vehicle lumbering back to home
base), remembering where each target has stopped and, most
importantly, deciding when to send paint down-range. Throw in the
good ole "friendly fire" problem and you've got yourself a nice little
Artificial Intelligence problem.
But, one last time, for argument sake, let's assume that all of the
problems above can be solved. Then, your million dollar solution just
has one last scenario to deal with: The opposing operator waves their
hand while their vehicle is stationary until your marker is out of
paint. And, if you solve that one, the operator sticks a 12" photo of
a tank into the ground on a spring and starts it rocking back and
forth.
For all of these reasons, the "The Pittelli X-Prize" (as Doug has
named it) is hereby officially announced:
"I, Frank Pittelli, will pay $1,000 (US) to the first person who
successfully deploys an autonomous vehicle on the battlefield that
scores a single hit on a designated opposing tank before being
destroyed itself during a normal R/C tank battle."
If anyone thinks the challenge is too easy ... feel free to take my
money. If anyone feels that the challenge is too hard, feel free to
explain your reasoning and ask to make it easier. Having watched
robotic challenges for many years now, I understand the need for the
organizers to make them unrealistic and simple enough so that the
roboteers will accept the challenge :-)
Frank "Autonomous Robots Can't Handle The Real World" Pittelli
On May 17, 6:20 pm, "Doug Conn" <[email protected]> wrote:
> You have to see this video. These guys are contenders for the Pitelli
> X-Prize.
>
> http://www.youtube.com/watch?v=RxBa5bQfTGc
> <http://www.youtube.com/watch?v=RxBa5bQfTGc&eurl=http%3A%2F%2Fwww%2Efu...
> ll%2Ecom%2Fmain%2Ephp%3Fmod%3Dvideogallery&feature=player_embedded>
> &eurl=http%3A%2F%2Fwww%2Efutureball%2Ecom%2Fmain%2Ephp%3Fmod%3Dvideogallery&
> feature=player_embedded
>
> - Doug
--~--~---------~--~----~------------~-------~--~----~
You are currently subscribed to the "R/C Tank Combat" group.
To post a message, send email to [email protected]
To unsubscribe, send email to [email protected]
Visit the group at http://groups.google.com/group/rctankcombat
-~----------~----~----~----~------~----~------~--~---
