Boslough, Mark B writes:
> I like this. I am using evolutionary computing methods to modify flight
> behaviors, but they could also be applied to optimize aircraft models for
> whatever property that you want to optimize. The method is based on
> biological evolution (survival of the fittest and all that). This is how
> birds learned to fly and tuned themselves to their environment (why
> albatrosses have a different aerodynamic form and set of behaviors than
> buzzards, for example). With JSBSim you can parallelize and run simulations
> off-line on a supercomputer (no visualization, just the FDM), and then
> extract the optimized values. Anybody want to help me with this?
Not that I have extra time for anything new, but here's what I did for
my grad school project (and yes, I paid my $1 so I am authorized to
use the "cool it works with Linux" logo, but I'm probably dating
myself there, although that may not be legal yet in all 50 states.)
http://www.menet.umn.edu/~curt/research/
We didn't call it an "evolutionary method" though, we just made a WAG
and tested to see if we were any better off than before. :-)
Essentially our *very large* search space was partitioned
"statistically" which means if every process is off making WAG's,
their's a good chance they each are exploring different portions of
the overall search space. This sounds pretty hokey at first, but
actually works surprisingly well.
Each process was running a self contained AI search algorithm and was
oblivious to what all the other processes were doing.
The first process to find a solution "won". The assumption here is
that the most quickly found solution is probably the shortest and most
optimal. For other applications you might need a different metric
... i.e. the process that hits the ground last wins ...
I was ganging up 25 linux machines to find solutions. Earlier
versions were developed using a honest to goodness, real life super
computer, but it was a pain in the butt to get time on it, and
everything was batched and ran lower priority than the paying
customers, so it was usually faster to run on your own PC(s). A nice
feature of what I was doing was that none of the processes needed to
communicate with each other other than to call "BINGO!".
I'd be happy to answer more specific questions about what I did, and
you're welcome to browse the code and help yourself to any of it.
Unless otherwise stated, assume GPL ...
Curt.
--
Curtis Olson IVLab / HumanFIRST Program FlightGear Project
Twin Cities [EMAIL PROTECTED] [EMAIL PROTECTED]
Minnesota http://www.menet.umn.edu/~curt http://www.flightgear.org
_______________________________________________
Flightgear-devel mailing list
[EMAIL PROTECTED]
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
