[ODE] Yup! I have that problem exactly - saving virtual creatures

[Robert Leclerc]

The differences are not THAT big, but it could be big enough to make a
difference. That is, since this kind of evolution is basically a gradient
climb, you could get stuck by having a genotype produce better fitness
than it is "really" going to give, and never do better than that. Thus all
subsequent genotypes will be converging on this one. I was running my
simulation and I think I was getting that kind of behvior.

I think we can rule out percision errors on saving being the problem since
if you just use the same NN values and reload the body (without exiting
the simulation) you still get differences each time.

Tomorrow I will try destroying all the space and world stuff and see how
that goes.

Best,
RDL


> How big of a difference are we talking here?  Invisible rounding error
> type stuff or errors big enough to notice visually?  I'll take a look
> at his code after I get my stepfast stuff integrated into the CVS and
> the documentation written, hopefully tomorrow.
>
> Try destroying and creating a brand new world, new ground plane,
> everything, each time you reset the simulation.  Maybe that will help.
>
> David
>
>> I got same problems about save-load virtual creatures.
>>
>> In my system, follows karl's approach, the genotype is converted into
>> phenotype, then physical structure is built by parsing the grown
>> phenotype digraph.
>>
>> I saved the genotype information in the file, but when I loaded and
>> simulated again, the fitness was slightly different.
>>
>> Unfortunately, even when I run the simulation twice with same digraph
>> structure(not in file, in memory), the two result is also different!
>>
>> What's wrong with ODE?
>>
>>
>> <center>Free webmail hosting service MNARA.NET<br>-- This e-mail was
>> sent by FREE KEBI Mail at http://kebi.com/ --</center>
>
>
>
> _______________________________________________
> ODE mailing list
> ODE@q12.org
> http://q12.org/mailman/listinfo/ode