Subject:
|
Re: Design
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Thu, 8 Dec 2005 15:05:57 GMT
|
Viewed:
|
1398 times
|
| |
|
|
I do remember seeign something very similar to this during my studies,
except it was a genetic alogorithm based on rounds in an arena. The
genomes would gain fitness by being able to push their opponent out of
the ring. Some of the most sucessful were just vast blocks which would
just fall over.
There was also a recent study into flying, with genetic algorythms
evolving the most effective flying wing pattern. They reached a point
where some were merely standing on wingtips to get higher than less
sucessful attempts. Others used the wings to jump.
This exploitation became even more apparent when microcontroller/FPGA
based firmware solutions were evolved using a GA and a real test rig.
The sucessful genomes wouldnt translate to a similar (but different)
model and run. As it turned out, after some serious root analysis,
they had exploited the capacititive properties of the proximity of the
data lines on the particular board they were using to acheive part of
their task - this was unexpected, and a pattern a human designer may
never have considered.
I find the very idea that genetic algorithms will abuse and find
loopholes in their envirnoment fascinating, and just try to think
where it can be exploited. On the other hand, there is nothing to stop
taking a cross section of a simulated population, and tighten up the
simulation around them as they go - thus forcing them to adapt to a
changed environment. In this respect - a multi-platform grid based
simulation would offer something in that it is not running on symetric
platforms, and sucessful genomes would have to take this into account.
I certainly feel that something derived from GA's, serious physics
simulations including electronics and fluidics mixed with an
environement of bricks with known parameters, could evolve something
very useful. Push it in the direction of clustered robots or motes and
the reuslt is a very adaptable system.
Has anyone yet tried to replicate the sucess of the "reproducing"
blocks from one particular university with Lego? If not, why not? I
may have a shot at that if I have a chance to do so.
--
Danny Staple MBCS
OrionRobots
http://orionrobots.co.uk
(Full contact details available through website)
On 07/12/05, steve <lego-robotics@crynwr.com> wrote:
> PeterBalch wrote:
>
> > I love it! Is it really true? Do you have a reference?
>
> I first saw this in a SigGraph course many years ago.
>
> Googling around, I'm fairly sure this is the guy who did the work:
>
> http://www.genarts.com/karl/evolved-virtual-creatures.html
>
> He's come a long way since those early days - his 1994 paper describes
> an approach to preventing the 'falling over' trick.
>
|
|
Message has 1 Reply:
 | | Re: Design
|
| (...) Another story like that - relating to neural networks - is of a busy Japanese subway system where they wanted to prevent overcrowding of platforms at the stations. The story goes that they planned to hook up a camera with a computer observing (...) (19 years ago, 8-Dec-05, to lugnet.robotics)
|
Message is in Reply To:
| | Re: Design
|
| (...) I first saw this in a SigGraph course many years ago. Googling around, I'm fairly sure this is the guy who did the work: (URL) come a long way since those early days - his 1994 paper describes an approach to preventing the 'falling over' (...) (19 years ago, 7-Dec-05, to lugnet.robotics)
|
6 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
|
|
|
|
