Darwinbots Forum
Code center => Suggestions => Topic started by: Zelos on August 20, 2005, 03:56:16 PM
-
Proofread by Numsgil
I have a few suggestions. I think, in later versions after the astronomy part I posted about before is done, we might add something so that bots can only see specefic parts of the spectrum. "Why?", you are probably wondering.
1. One reason is that it gives more options for evolution. Like if a bot is poisonous and has the color red in its dna (it says red as start color, but the color we see doesn't change) then others can evolve to think red=poison and then avoid bots who are red.
Of course then other bots that aren't poisonous might become red. This can also be used for identification. A species can have a specefic color and use it to identefy other members. With this, illusion and deception is possible. A species can evolve to get that color, sneak in and attack bots and they wont know what hit them. There are alot of advantages.
2. Bioluminescence (sp?). 100-3000m below the ovean surface many animals are bioluminescent. Why not add it to DB? If we have light filter through water (which already exists in pond mode) all the light will disapear after a certain depth.
In such environments bots might either learn to create their own light (bioluminescence) or navigate without any light at all. Eyes can still have a wavelength parameter which shows it at what range it can see. But we could also add that different wave lenght can go different distances in the water.
In reality is it so that at those deepth red cant go very far but blue can so most animals only see blue and green, not red. so a jellyfish have a red glowing stomach to neutrolize the blue color of its pray so it becomes invisible (for other fishes).
Bioluminescence can also be a way to communicate. With some clever programming, bioluminescence can be used to yell "come here guys, food is served" and then all the bots go there. There's alot of potential here.
What do you think?
-
I think the first idea is sort of already done; the ref variables allow a bit to know whether another bot has any poison commands in its DNA, even if they are not expressed to create poison. So in that sense, a bot can evolve to pretend to be poisonous even if it isn't. But a poisonous bot can't pretend to be non-poisonous I suppose.
As for bioluminescence, I suppose the in/out variables could be modified so that they can be viewed at different differences depending on the depth of the bot.
-
dont you think it shall be more biological than artificial?
-
I agree with Ulcisor that alot of this is in the program already. the only thing that isn't present is decreased visibility as the light level decreases.
-
Do you think that would be a cool idea to add, [Num]? I can't think of applications that would really need this but it would be cool to add anyway. I suppose a bot could be programmed to go below a certain depth when being attacked to escape the attacker, or could always stay at a certain depth to allow it to sneak up on prey, so to speak.
-
it sure exist but its to artificial if you ask me. this is supposed to simulate real life as good as posssible, shouldnt we then remove as much artificial as possible and replace it with something more biological correct? I dont think a mamal can see at a snake and see the genes that tellsits poisones. If the snake is red/yellow its almost certain to be poison.
-
Isn't it basically the same thing anyway though?
A mammal sees a reptile and knows it is poisonous because of the colour.
A bot sees another bot and knows it is poisonous because of the reference value. The bots don't see the genes that contain the poison command, just the fact that the commands are there.
-
but the thing is that the value is not choiceble. like if a bot "want" to go out and be eaten and are poisonus its not possible for it to be eaten cause the other bot sees something that the poison bot dont want to show. to see how many commands you have is like for a lion to look at a prey and see its dna. its not possible, but it can see its color and that can tell things.
-
Actually the ref values are extremely unnatural. I dont have a better solution at the movement, but still it seems to me in one way or another that zelos' suggestion is one of the better, though it still needs some refinement.
-
I have advoocated for a specific genotype and phenotype before. In reality, what we see are the phenotype, which is the expression of the genotype.
-
of course our color shows the gene, its the genes that makes it that color. but what kind of refinment do you suggest? if we have the normal eye that already exist. a gene that tells what range of frequence or wavelenght it can see. and then all colors below or above it becomes invisible. (of course all object reflect light of all wavelenghts but the invisibility comes clear in areas where light dont reach such as at deep oceans). the range of seeing can be like enzymes coded at the end of the genome and cant be changed over the life time. And the bigger spectrum you see the more costly it is to use commands connected with the sight. This prevent from all seeing bots.
-
What I mean that your idea should be more refined, were it was only general in description, and not very detailed, you know discussing various methods etc.
-
About the all seeing bots, I know that bots dont have a temperatur, but in principle it should be possible to see infra-red and thereby see all things. We could also say ultra-violet, or because "sight" aint resticted by the spectrum we see, maybe a 'Gamma'-bot can evolve ;)
-
thats the thing, we can use a prett wide spectrum, not the one we know. but gamma is mutating so its out of question. but IR-UV is very possible. we could make it look like this at the end of the dna:
cond
start
stop
sight
300
700
endthis mean it can see from 300 terahertz to 700 terahertz. then we can set up a forumla that decied how much it cost to have that sight per cycle. like the square of the differens or something so it becomes more and more expensive for each extra hertz it sees. This would prevent from all seeing bots that sees everything.
But what about the reflection? the same there but it would look like: color
350
159
560this bot reflect all those colors. but the color should be able to change under the life time with some comand like .color so we can get camelonts. that change color depended on things. But it would be quite expensive to change and take some time. The colors you start with might cost the square root of the amount of colors while changing ur color might cost the cube of the amount of colors you already have.
and with Bioluminescence it could look be another part at the end that tells what colors you can lumonate. the more chemicals you have to luminate the mroe expensive is it to just stay alive. it could look like:Bioluminescence
750
680
126
257all those frquenses could it lumonate with this command:750 .lumo storeand voila it begins cloving, of course this require energy. this lumo last for a cycle so it will have to keep lumo unitel it things its enough.
how about this ideas?
-
Yet again...: Strong colors (like red/yellow) often indicate that an animal is poisonous, with snakes often have a more neutral color. I also think of the shape and smell of an animal plays an important role. Mice learn that snakes and cats are there main enemy, it is not coded in the genes, that they should avoid, a flat, long, hissing, animal with x .poision genes in it... Not saying that we need to get a learning algorithm in the bots, maybe we can come up with other things.
-
Zelos, this looks good to me. But I would be afraid to letting more commands get into DB, because we already have to many. And because of all this ref-value thing, I think need revision, it may have to wait or get more refined. I dunno
-
if we count in that smell is coming up later 26 ref commands wuld be replaced with sight and smell wich togather is only 2 commands. so its a lose/gain of 24 commands. but it would be somethings at the end of the DNA. with that we would get 5 commands so the gain is then 21, so with this system we acctualy would remove commands and gain more realistic ones. isnt that what you want?
-
Yes that is what I want. Fewer commands, that dont need operands, is making evolution faster (this have been proven troughtout the entire AL literature.) I am on the a total revision of DB. I know all the old bots gets incompatible with the new one, but with clever programming and not too complaining Game programmers, it could be possible to get a DB people will like much more and getting one that is more realistic. Not that we need a 100 % to nature DB, but one that doesnt impose strange and artificial way for the bots to behave and get strange information. Right now it is extremely difficult to analyze an evolved bot. This has nothing to do with that DB is "complex", but that we have a million different commands, and that bots are getting unrealitisc much information at the moment. And that a lot of the commands, need additional operands to have a meaning, making evolution slower than it could be, and thereby building up junk DNA or what we currently at the moment think is junk DNA.
I once had a bot text file, with over 4000 lines...........? Hell yea it would take a month to analyze that bot. And when analyzed, it is often that 99 % of all mutations are junk/intron etc., and the basic parts of the bot are those that the original bot had.
-
Do you think that would be a cool idea to add, [Num]? I can't think of applications that would really need this but it would be cool to add anyway. I suppose a bot could be programmed to go below a certain depth when being attacked to escape the attacker, or could always stay at a certain depth to allow it to sneak up on prey, so to speak.
Yeah, and it wouldn't even be terribly difficult (sort of. Being high and looking down might be a bit more involved if you want vision to be obscured by where you're looking instead of where you are. And if that didn't make sense, now you know why it's an aside). But it should probably be tied closely with pond mode.
-
If someone would go trhough all the sysvars (in sysvars2.21.txt) and figure out what commands can be combined, eliminated, etc. We could probably go from there.
For instance, with the way DNA will work very soon, we can combine fixpos and fixed into a single read/write value.