Slashdot Mirror

← Back to Stories (view on slashdot.org)

Tom's Looks at Two DARPA Grand Challengers

Posted by timothy on from the thrum-thrum-thrum dept.

skeeball writes "As a follow-up to this article, Tom's Hardware has a behind the scenes article on two of the teams competing in the DARPA Grand Challenge 2005. "The Defense Advanced Research Projects Agency (DARPA) hosted the first Grand Challenge Project last year, offering a reward of $1 million. This year, the prize money has been doubled, making the competition all the more interesting.""

3 of 169 comments (clear)

  1. Article Link by Kozz · · Score: 5, Informative

    Too bad the submitter didn't Link the Article itself.

    --
    I only post comments when someone on the internet is wrong.
  2. That's so Tom's Hardware by Animats · · Score: 5, Informative
    That's so Tom's Hardware. "7 Pentium M CPUs!", and no word about the algorithms. They could have at least said more about the sensors. Actually, everybody's sensors suck. The radars can't profile terrain, the LIDAR units are only line scanners, the stereo vision systems have trouble locking up on dirt, and the vision systems are a long way from being intelligent. True 3D LIDAR is coming, but not this year. The Grand Challenge rules prohibit the use of the best available 3D LIDAR system, because it was developed with Government funding and wasn't available by August of last year.

    So we have a line-scanning LIDAR on a tilt head, like CMU, which is an adequate but bulky solution..

    We have two industrial Pentium 4 machines running QNX, on our Grand Challenge entry, along with five Galil programmable motor controllers. We have room for 3 CPUs, but the compute load fit on two of them, so we took the third one out.

    Technically, QNX was an excellent choice, but because few people know it and many don't want to learn it, using it has made recruiting difficult.

    1. Re:That's so Tom's Hardware by Animats · · Score: 5, Informative
      Why is the vision processing so poor?

      Because, despite decades of work, vision processing of unstructured scenes still sucks.

      There are things that work in computer vision. You can do stereo, if the image has strong edges in it. You can pick out big moving objects. You can find the horizon. You can work out your own positional movements from video. You can find faces, align, and recognize them, sort of. You can find known objects in any orientation (which is very useful in industrial systems.) You can follow roads.

      Beyond that, not much works.