| |
On Tue, 23 May 2000, Ciaran Jessup wrote:
> - Anybody out there got any experience with Genetic Algorithms in LegOS,
> anybody willing to give me a little help in getting a handle on it, I'm
> considering a final year CSSE project using the RCX, and would be interested
> to here from anyone who's had experience in this
I've pondered it, though not actually done anything. The things to
remember are that GAs need to do fitness testing, and the larger the
population that you can do fitness tests on the better. On a big old PC,
it's easy to do lots and lots of fitness testing, since you can calculate
to your heart's content these days. In contrast, if you are doing pure
calculations to determine fitness, then the RCX has a slow processor,
which means it will take longer to test than it would elsewhere. If you
want to test against the "real world," then you are talking lots and lots
of time to do testing. What that boils down to (IMNSHO): if you want to do
GA on the RCX, you have to pick simple genes or simple problems, so that
you have to do less testing or can do more efficient testing.
Just my two cents, from having looked at the problem quite a bit...
Luis (who really hopes that he is proven wrong)
P.S. You could always compute on the PC and get 1/2 dozen robots to run
around "testing" the genes... that would help ;)
-----------------------------------------------------------------------
"Going to California with an aching in my heart.
Someone told me there's a girl out there
with love in her eyes and flowers in her hair."
-Led Zeppelin, "Going to California"
-----------------------------------------------------------------------
|
|
Message has 3 Replies:
 | | Re: GA in RCX
|
| (...) Yes. Therefore you might want to use the truncation selection, which allows you to have only 1 copy of the population at a time to save memory. It worked for us, we have a small GA running inside of the RCX. We evolved a program with simple (...) (25 years ago, 24-May-00, to lugnet.robotics.rcx.legos)
| | | Re: GA in RCX
|
| (...) I like the multiple robot idea. If each robot used a simulated annealing algorithm instead of a GA this would require less computation since it would be working on one chromosome instead of a population. Each robot could advertise its fitness (...) (25 years ago, 25-May-00, to lugnet.robotics.rcx.legos)
|
Message is in Reply To:
| | GA in RCX
|
| - Anybody out there got any experience with Genetic Algorithms in LegOS, anybody willing to give me a little help in getting a handle on it, I'm considering a final year CSSE project using the RCX, and would be interested to here from anyone who's (...) (25 years ago, 23-May-00, to lugnet.robotics.rcx.legos)
|
8 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
|
|
|
|
