Darwinbots Forum
General => Leagues => Bot Challenges => Topic started by: Elite on September 24, 2007, 01:41:03 PM
-
I've come to percieve a certain flaw in the way we program bots at the moment: bot behavior is generally too simple. Most bots' behavior comes straight from a series of if ... then triggers, directly from the dna; bots ignore others of their own species, and spend all their time eating (or searching for something to eat) and replicating. We're programing more and more efficient grey goo.
The purpose of these new challenges is to encourage the development of more interesting, interactive and intelligent bot behavior. The new challenges are more informal too and less competitive, with much looser rules and goals. Discussion and sharing of ideas is also encouraged.
This will also have consequences for the DB environment. Terrirorial bots will diversify the field; bots that can learn can utilise nontrivial strategies; bots with behavior controled by meta-systems (ie. non-genetically-deterministic behavior) might evolve in different ways ...
The first of the new challenges (http://www.darwinbots.com/Forum/index.php?showtopic=2235) is to program a bot with a very basic artificial neural net to govern behavior. To attempt this you need to know a bit about neural networks, so it's a pretty high target. Massive kudos to whoever attempts this.
-
i agree, elite, but its going to be a very high mountain to climb judging from the way we are programming at the moment... but over time, im sure we will rise to the challenge :)
-
I'm making a bot that has the strict rules of the if then triggers but generationaly adapts to the environment without mutations. In fact it would degrade over time with it. It over time decides what to use and has the ability to change to adapt to almost any opponent. Although i only have a small part done :( but this is the next step to a neural net type bot.
-
In fact it would degrade over time with it.
I don't really buy that, any bot can benefit from mutations under certain circumstances.
But, good luck with your first big bot project :)
-
this bot has the ability to use everything from all sinces and any mutation would shatter the uses of them and i made a bot that i used as a base and over night it had almost no mutations and some had none. the only mutations it had were ones that had no effect to the other genes like insertions at the end or middle of genes. i was sad that so little had took place, but it did what it was suppose to do. :)
-
Should be an interesting bot;
One good way to see if mutations improve a bot (now this really does not work all the time)
a.) Make to copys of your robot, lets call one bfb and the other mfb (that is the names I use in my runs)
b.) Start a new simulation with both robots and some vegys. Make sure to set mfb to reasonably high mutations and make sure to completely disable mutations for bfb.
At some point you should see the mutating robot (mfb) kill off the non mutating robot.
(I have some tricks to make this process work consistently, I just did not implement them yet)
-
my first neural bot is done and is very interesting 8)
still waiting on a good useful bot though
dbv 2.46
you may have to disable each individual mutation though because of mrepro
enjoy and if you wish to modify it making it better i'd love to see
-
Does the NN estimate itself over the robots life time?
Also, is the robot passing its NN settings to the child?
Another thing that will help is to add some basic self awareness rules.
For example: Children are naturally afraid of heights or afraid of the dark.
Our neural networks build around the hard underling rules we emotionally inherit from early childhood. Fear is one possible emotion you may work with.
-
currently i've got the NNet mutations in place but i can't figure out how to pass the parent NNet to the child properly. i'm wanting to do ties to pass information through. it's becoming a mess so i'm trying to improve it the best i can and enable it to have more mobility so it can do more than one thing at once. what i have so far is in the attachment if you want to help.
-
small update
-
working on an explanation of what i want in the .txt should have it by tomorrow
i haven't worked on it in a while
EDIT: never mind, it took a lot less time then i thought
the explained bot is in the attachment and any work done on it please post so i can see and thank you if you do :)
EDIT: all bots with .x (now 1.5) are not complete usable bots they are only to be worked on and not ran in sims yet if anyone else is wandering
-
i've condensed the NNet bot to only use radical mem and limited it's ability however it is working perfectly. :D
EDIT:ok i've just found a problem and it's not with the bot code it's skipping memlocs and shouldn't be.
-
i fixed the issue the bot dna works fine now (- radical mem at gen 2 for me).
-
I have not forgotten about you Shadowgod2, I just can't work on this stuff Mondays and Thursdays.
I am making some good progress fixing the rest of epigenetic memory.
I should have a drop for you today.
Sorry for the wait.
-
actually i think that was fast thank you :D
-
Neural networks, We need more systems for bots to communicate...Give us more sensors dudes.
Anyway, more diverse bots, thats like...My current project (Which im hoping moonfisher will help on)
Im scrapping the, if 1 condition met, the whole task is done...Im making it so each task is about 4 to 10 genes that fire to perform it, (Gonna need some custom-sysvars in the empty-sysvar block to check which genes already fired) then the task slowly gets completed (or more quickly actually...) with options from about 25 genes, to 40 genes to select from...
Of course its going to only perform a few tasks, but they'll be good tasks.
-
well this thread hasn't had anything in a while.
ok so i've been playing with Nnets again with an idea of making it as real of an actual neuron as possible and it has the most potential as i've seen.
although many will not move or react to anything many will act differently at different times as just of a matter of another bot moving or still(same bot triggering the reaction) i will have to develop this further for sure. you'll just have to see it your self.
EDIT: also to see the effects easier i recommend thin fluids.
also i think i found a bug so where it freaks out when adding new bots to the sim while it's playing so add them while the sim is paused.
-
got a fix in the dna it basically skipped age 0 and couldn't figure out what to do.
-
I do not understand the business end of a NN as shown by Ann. But I do know enough that the Neurons should reorganize faster with more pain and reorganize slower with more pleasure. This should lead to a stable supper efficient robot after a while. Just my 2c$.
-
yea i'm definitely messing with pain pleas this time and going from there
-
yea i'm definitely messing with pain pleas this time and going from there
Hey shadow, may I make 1 request in your neural nets.
If were gonna have a kinda realistic neural net, don't make ties between each neuron. Use infoshots or some similar system like the synapses in a really nnet.
Maybe we could get a long-time simulation for it running then, a bit like Botsareus' zerobots :P
-
offtopic: sorry for the silence i was on vacation sense Saturday just got back.
it's not that kind of Nnet. just like the Nnets before it's with a single bot. each neuron of it's Nnet is as realistic as can be in as short of a line as possible. before i can do anything like that i would need a lot more experience in making bots multibots for sure and with my iffy go in this it'll take many more years to get to that lvl. i still struggle with the advanced stuff like >>. to make bots with them is a nightmare to figure out how they work exactly.