Subject:
|
Re: Design
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Sun, 4 Dec 2005 21:58:32 GMT
|
Original-From:
|
steve <sjbaker1@*stopspam*airmail.net>
|
Viewed:
|
1259 times
|
| |
|
|
dan miller wrote:
> > > I can think of one field where autonomous robot-like, pseudo-AI is alive
> > > & kicking, and making tons of money. One word: videogames.
> > > You click on an object, and the character picks it up.
> > > These are very sophisticated actions,
> >
> > Toy world!
>
> Hmm, yes in a sense. Why the negativity about constrained or simulated
> worlds? We are not Nature; we are not constrained to work in a design space
> that never changes.
>
> You can mean different things by calling something a 'toy world'. I would
> propose that it would be much more difficult to design a functional robot in
> a complex, unconstrained 3D simulation, than it is to design something like
> the Roomba, which works in the 'real world', but in a very limited domain.
To get to the goal of autonomous and 'sufficiently intelligent' robots,
we have a lot of problems to solve. Some of these are to do with
mechanics, battery technology, communications and sensors - others are
to do with solving problems, path finding, goal seeking and such.
I think the way that robotics develoment and video games AI is
proceeding is entirely complementary.
In the video game world, we are relieved of the problems of sensors,
mechanics, battery and such - so we can concentrate on solving the
tricky issues of of path finding, goal seeking and problem solving.
In the robotics world, we are a long way from being able to solve those
issues because we're still working on the lower level issues of how to
sense the world, how to negotiate tricky obstacles and that kind of
thing.
Rooba lives in a 'world' that's far more idealised than the video
game world. The ground is flat, the obstacles can be largely considered
to be 2D in nature and the goals are trivially simple (drive everywhere
once). Yet it's still a totally inadequate solution - Rooba gets stuck
a LOT - it can spend all of it's battery power navigating a forest of
chair legs under your dining room table - and miss huge areas of easy
carpet just a few feet away. Video games characters have a much
harsher environment to deal with - and their AI rarely gets them
stuck like that.
However, the Roomba has to deal with real issues of battery life,
motor stalling, carpets with tassles on the ends, your cat,
user-friendly controls to indicate regions where cleaning is not
required, etc.
But there are many elegant algorithms that have come from the games
industry (A-star navigation, Potentially Visible Sets, Adaptive social
behaviors, etc). These are things that robotics will eventually
benefit from once all of the low level issues are solved.
Saying that video game developments are irrelevent because they aren't
dealing with the real world is ridiculous. Development can be started
from both ends of the problem spectrum and meet somewhere in the middle.
It cuts both ways though. My own OpenSource 'tuxkart' game uses
the important bits of a line following algorithm that I wrote for a
Lego Robot.
> > I would love to put cheap autonomous toy robots for into the hands of
> > children (and adults) who might come up with the wacky ideas that will
> > actually work. The intellectual attack seems to be getting us nowhere;
> > maybe evolution will work - it has done before.
I strongly disagree that the intellectual attack is getting us nowhere.
We have real-world (and somewhat useful) robots such as roomba and aibo
and we didn't have them 10 years ago. That's a success. Fields like
speech recognition are now so routinely used that you can phone an
airline booking service and TALK TO A COMPUTER. We forget that
speaker-independent single word recognition was impossible 15 years
ago.
The intellectual attack has been disappointing for several reasons:
* Firstly because the problem initially looked a lot simpler than it
finally turned out to be. The progress so far has been in
discovering just how hard the problem truly is. That's not a
lack of progress - that's a necessary first step.
* As soon as a computer can do something (like voice recognition
or playing chess at grand master level), the general public
mentally remove that from the field of AI. People naturally
define the word 'Intelligence' as the thing that humans can do
that computers and animals can't do. Look in an AI
textbook from the 1960's and you'll see playing chess at
grand master levels was one of the goals of AI - it was right
up there with the Turing test. Look in a biology textbook from
the same era and you'll see that 'tool using' is unique to human
intelligence - when we discovered that blackbirds use tools, the
definition changed to 'making and using tools' - then we found
monkeys that make tools - so then they switched to 'passing on
knowledge of tools to offspring' - and now we know that chimps
do that.
* We tend to trivialise some applications of AI. Because the only
use we have found for pathfinding in AI has been in video games
and robot vacuum cleaners - we don't see that as a significant
advance - but there is some sophisticated technology going on
'under the hood' of many video-games. Just try writing one and
you'll soon find out how hard it is!
|
|
Message has 1 Reply:
 | | Re: Design
|
| (...) Agreed. ... (...) One thing that game worlds give you that is very tricky in real-world situations is, the breakdown of the environment into separable objects. Game code starts with a bunch of objects (floor, walls, tables, (...) (19 years ago, 4-Dec-05, to lugnet.robotics)
|
Message is in Reply To:
| | Re: Design
|
| (...) Hmm, yes in a sense. Why the negativity about constrained or simulated worlds? We are not Nature; we are not constrained to work in a design space that never changes. You can mean different things by calling something a 'toy world'. I would (...) (19 years ago, 4-Dec-05, to lugnet.robotics)
|
7 Messages in This Thread:
  
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
