[ODE] First phase of program finished.

[Mattias Fagerlund]

> I've gotten the first phase of my genetic algorithm program done.

[snip]

Cool stuff, looking forward to seeing what you evolve!

> The second simulation, with 37 bodies and 36 joints (32 hinges, 4
> fixed), just barely keeps up with realtime.  

[snip]

The obvious test of where the CPU cycles go would be to turn of the GL 
calls all together, and comparing the fps to what you get running with 
GL.

One performance increasing tip from my own evolving sessions; set an 
early death limit that states that if the agent doesn't perform X within 
the first Y seconds, it's bad and should be terminated. This means that 
effort isn't wasted on the first sub-par performing individuals. For 
instance, if the measure is how far an individual travels in 15 
simulated seconds, kill all who travel less than 3 meters in the first 4 
seconds. Set their performance to how far they actually travelled during 
those 4 seconds, so there's a fitness gradient within the early 
generations where all fail.

In case you haven't seen it before, I have some ODE evolution here; 
http://www.cambrianlabs.com/mattias/ and Gene Ruebsamen has some ODE 
evolution here; http://www.erachampion.com/ai/ .

cheers,
m