Bots and Simulations > Simulation Emporium
Q: shortest zerobot reproduction time
ikke:
--- Quote from: bacillus ---How do you apply evolutionary pressure to force reproduction?
--- End quote ---
Evolutionary pressure forces reproduction speed The lower edge is determined by minimum body / energy requirements and the upper by ageing cost and / or degradation by point mutations.
--- Quote from: EricL ---In fact, reproduction is actually much easier to evolve then feeding via .shoot. It's quite a sophisiticated thing for there to be sufficient selection pressure to favor .shoot. I've never seen it occur in zerobot sims. There has to be significant benifit for it to be favorred and selected for and without aiming, which requires vision, which is very sophisitcated, random shooting doesn't tend to convey enough benifit to be selected for in my experience. At least not yet.
--- End quote ---
Hmm, looks like I'm in for the long haul. I have a sim set up trying to do this. I evolved replicators as veggies and am now trying to evolve a saved version into a feeder. I have costX negative to sustain them and a field full of veggies for lunch. You never got this working??
EricL:
--- Quote from: ikke ---You never got this working??
--- End quote ---
In a word, no. No one to my knowledge has ever evolved anything as complex as a bot that can hunt or see, starting from a zerobot. There is even some uncertainty as to wether anyone has ever evolved a zerobot that deliberatly utilizes conditional logic to make a decision. Trafalger as put in for the Conditional Logic prize and I have been too lazy to review his entry. My bad. He gets the prize no matter what, I just need to find an hour or two to rub together to walk his entry.
This said, most evo sims were run prior to the in-line conditional DNA logic change, which should make evolving conditional logic more likely. Additionally, the use of self-throttling shepards and preditors which control their hunting effectiveness via .totalmyspecies and .totalbots is a relatively new invention and one which IMHO, holds great promise for applying selection pressure in such a way as to directly favor the evolution of complex behaviour.
If you want my $0.02 on how to best accelerate the evolution of complex behaviour, I would recommend against using negative costX because it is sim wide and does not favor individuals. Better is to use veggy shepard bots that shoot nrg shots. In this way, bots can evolve to aquire more nrg individually by stopping near a shepard, following it, increasing their body and thus their cross section, increasing the probability of a nrg shot hitting them etc. Favorable adaptations in individuals are selected for. Have the shepards target the bots with good accuracy at first, even follow them around if you want, but have them get more and more random (or shoot less pwoerful shots, or less often, or move away, etc.) as populations grow and the evolving bots gain effectiveness. Writing the right shepard for the right evolutionay stage of your sim is the best way IMHO to apply the right selection pressure at the right stage of evolution to drive evolution towards complexity. Eventually, the shepards can turn into preditors that prey on the bots, etc. Hopefully someday, we will have what we see with evolved bots that began as hand coded bots in Internet Mode and that is that the evolved bots are very very hard to beat with hand authorred code. Hopefully there comes a day when you can't write a preditor your evolved bots can't eat.
Oh, and I shoudl correct myslef above. I have seen selection favor bots that constantly shoot feeding shots in sims with zero costs. The probability of them bumping into something and gettind fed is sufficient to favor this adaptation. What I have never seen is active targetting or conditional shooting when something is in view or comes into contact. Unconditional shooting of feeding shots is on par with unconditional reproduction w.r.t. the ease of evolving it from a zerobot.
Endy:
Just wanted to say that you can get alot of the same effects from waste build-up. Reproduction and share-feeding(w/ties) are the easiest to get this way. Probably similar to what happened in nature, a chemical process promoting replication prior to a self-replicator appearing.
The bots do seem to favor either continous or periodic actions over normal conditional logic. I think most of has to do with the problems associated with conditions breaking down from mutations, might be interesting to run a test to see.
ikke:
--- Quote from: EricL ---Oh, and I shoudl correct myslef above. I have seen selection favor bots that constantly shoot feeding shots in sims with zero costs. The probability of them bumping into something and gettind fed is sufficient to favor this adaptation. What I have never seen is active targetting or conditional shooting when something is in view or comes into contact. Unconditional shooting of feeding shots is on par with unconditional reproduction w.r.t. the ease of evolving it from a zerobot.
--- End quote ---
Okay, youre talking about shooters vs hunters. I'll settle for shooters right now and see if I can take it from there.
Numsgil:
Vision has gotten much simpler to use in newer versions since conditions can now be in-body. The simplest conditional feeder might look something like this:
.repro inc
*.eye5 50 >
.shoot dec
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version