Subject:
|
Re: an idea, can someone tell me if this is possible/been done before/etc?
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Tue, 2 Dec 2003 16:27:27 GMT
|
Original-From:
|
Gordon Elliott <GELLIOTT@CSISC.antispamCC>
|
Viewed:
|
881 times
|
| |
|
|
On "complexity", I would point out that the computational power of the
neural pathways of the house fly thoroughly exceeds the computational
throughput of our most massive supercomputer clusters at the moment.
True "adaptation" would require hardware functionally capable of learning
and feedback, and that can occur at various levels of complexity. All of
them will be light years away from the actual biological spider or house
fly, but interesting effects can be demonstrated with even simple equipment.
By the way, as humans we solve seemingly combinatorially complex problems in
fractions of a second all day long. Look around your house. Judging how
walls intersect to get an analysis of which are convex and which are concave
corners is an NP complete problem--yet we just look up and get the answer
immediately. (And there are optical illusions that take us time to settle on
one of several interpretations, and when we finally see such an
interpretation we have an "oh yea" moment as a consistent pattern emerges.)
The reason we can solve such problems quickly is that there are many clues
that allow us to resolve these in more or less linear time in most cases.
(So put me in the liner time group, I've only got one lifetime to do my
'computing.')
-- Gordon Elliott
----- Original Message -----
From: "Jim Choate" <ravage@einstein.ssz.com>
To: "pixel" <pixsrv@poczta.onet.pl>
Cc: <lego-robotics@crynwr.com>
Sent: Tuesday, December 02, 2003 8:24 AM
Subject: Re: an idea, can someone tell me if this is possible/been done
before/etc?
>
> On Tue, 2 Dec 2003, pixel wrote:
>
> >
> > "Kevin L. Clague" <kevin_clague@yahoo.com> wrote in message • news:Hp8uqD.8yM@lugnet.com...
> >
> > > I'd recommend pneumatics instead of large number of RCX's.
> >
> > but pneumatics gives you just automation
> > you dont have "adaptation"
>
> Neither does a RCX.
>
> > i can imagine sensor-like pneumatic valve configuration
> > but it would be much more complicated than using rcx (even few of them)
>
> No, it just moves the complexity to a different level. In the pneumatics
> it's in the hardware, in the RXC it's in the software. If you think this
> makes a serious whit of different with respect to the level of
> 'complexity' then you need to spend some more time learning what
> 'complexity' really is.
>
> I can't suggest Ilya Prigogine's "Order out of Chaos" (RIP buddy, we
> miss you). Further I'd suggest,
>
> Frontiers of Complexity
> P. Coveney, R. Highfield
> ISBN 0-449-90832-1
>
> I will say one thing about this one, be carefull about taking their
> examples with regard to the complexity of life with a grain of salt. Their
> discussion of why the Earth will never become a cold dead ball because
> photons fall on it is simply stupid and wrong. Otherwise It's a good book.
>
> > let's take a small obstacle into consideration
>
> Cool. Experiments (thought or otherwise) are the life blood of science and
> technology. I get so tired of telling people 'run the numbers'...nice to
> see somebody doing it ;)
>
> > when bot set the foot on the surface which is higher than expected
> > bot has to accomodate this obstacle
> > and decide if the original "program" of setting this foot (and next • feet)
> > will continued unchanged or some changes have to be made
> > (it will cause the bot's body will slope or not)
> >
> > solving this problem using pneumatics is VERY complicated
> > with rcx and even just touch sensors it looks (for me) much more simple
>
> I think this is a very complicated problem for a computer as well. As a
> start look into a technical term called 'compliance'. The 'step over'
> ability of legs is one of the greatest problems in robotics.
>
> Computing is computing. How it's done doesn't usually make a big
> difference (re O() notation algorithm complexity) except in run time. The
> problem here is that the run time is a factor of the solution and not the
> problem, mixing chickens and eggs.
>
> All computing problems that don't fall into the NP class (or higher) of
> complexity are all pretty much reducable to a universal Turing machine.
> And I have zero motive to get into the P==NP? question here, other than to
> say put me in the P<>NP! group from the get go. I see no reason to expect
> exponential or factorial problems to be reducable to a group of linear
> problems.
>
> Good luck.
...
|
|
Message has 1 Reply:
4 Messages in This Thread:
   
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
