Subject:
|
Re: Design
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Sat, 3 Dec 2005 13:17:36 GMT
|
Original-From:
|
PeterBalch <PeterBalch@=nospam=compuserve.com>
|
Viewed:
|
1339 times
|
| |
|
|
Bruce
> > Most people are happy if their robot can get from A to B without • getting
> > stuck behind the sofa. And that includes 99% of universities.
> > It's fun to speculate but it's a
> > futile daydream until we've got robots that actually work.
Tee hee! I seem to have stirred things up a little :-)
> In three years in the CS department of the Univ of Utah, my students
> were able to do much more than this using Lego mechanicals and JCX
> prototype hardware - autonomous checker and soccer playing bots, with
> some pretty advanced machine vision
These are extremely artificial environments. The problems are far easier to
solve than the "getting stuck behind the sofa" problem. (Ask any AI
researcher "how long will your autonomous robot suvive in my living room
without getting stuck?" and they'll start to look sheepish and try to
change the subject. The best researchers might claim a couple of hours.
Vacuum cleaners require nicely arranged rooms.)
My guess is that your machine vision for checker playing simply reconised
"black piece", "white piece", "no piece" and that the soccer playing bots
used an overhead camera. I'm sure your students are terribly intelligent
and put in a lot of hard work but these are toy problems.
> This team did a sonar mapping robot which was really cool:
> it scanned a sonar module like a radar antenna, divided its
> environment into grids, and filled in objects in the grid if there
> was a persistent "hard" echo from that portion of the grid.
Nice work, bright students. But did they go on to autonomously navigate
round the building? Did you have to have an artificially tidy environment
with no clutter on the floor and no overhanging tables or soft furnishings?
Too often, AI solutions only work in the toy worlds they were designed for.
Replying "but this was only a one semester project, what do you expect"
isn't a valid argument. Over and over again we have seen that AI solutions
in toy worlds simply do not scale up to the real world.
Even the Darpa Challenge was in a toy environment - the cars recognised
roads. If they went off the road, they died.
Not "getting stuck behind the sofa" is something that animals have been
solving successfully for half a billion years, often with far less
processing power that a PC. No current robot can do 1% as well as a
cockroach or an ant in terms of "not getting stuck behind the sofa".
It's apalling that AI robotics research has achieved so little in the last
40 years. New sciences usually achieve their most spectacular breakthoughs
when they're young but that simply hasn't happened in AI robotics. It isn't
as though AI needs lots of money. Physics or biology or astronomy have made
huge breakthoughs in that time even though they need a
billion-dollars-worth of telescope, particle-accelerator or DNA-sequencer.
With a bucket-full of lego and a desktop PC you can do cutting-edge
research. Which is great news for people with Lego but the AI researchers
should hang their heads in shame.
Peter
|
|
Message has 1 Reply:
 | | Re: Design
|
| Peter, Carrying on from my last post, you are right, AI research has done little despite its claims and researchers. The mighty cockroach still defeats the AI world. I don't think that is reason to despair, but personally, I look at it as a reason (...) (19 years ago, 3-Dec-05, to lugnet.robotics)
|
3 Messages in This Thread:
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
Active threads in Robotics
|
|
|
|
|
|
