Messages

31

Suggestions / Death Rate

« on: May 26, 2007, 07:41:17 AM »

I've just set up some sims using an age limit (massive instant death age cost becoming active at a certain age) and they indeed seem to work really well:

When I turned up the movement cost, the bots responded within a few hundred cycles and became noticably more conservative with their movement.
When I turned up the DNA upkeep cost, the average DNA length of the bots dropped rapidly. Within a few thousand cycles they had lost a gene or two, merged two other genes into one, and posessed a much more minimalistic genome.

What seems to be happening is that since living very long isn't an option anymore, the bots have to act in the best interests of their progeny or their DNA will die with them when the age cost gets them.

Nice find.

32

Bot Tavern / I miss SWARM

« on: March 10, 2007, 05:48:00 PM »

It was using come conditionless coding that depended on negative storing not functioning. Negative stores are absoluted now, so it was storing when it shouldn't.

33

Bot Tavern / I miss SWARM

« on: March 10, 2007, 01:40:10 PM »

I've released another version for 2.43. It's in The Bestiary

I've fixed it so all the coding now works, and took the oppertunity to use Eric's eyeXwidth command to give it 360 degree vision, which has really helped with the swarming. And now with an antivirus gene to give it a chance at F1 against virusbots without affecting it's usual behavior

I've attached a swarming vegetable: essentially SWARM 2.0 with all the attack genes stripped out and some minor modifications for collision avoidance and random reproduction staggering

34

F2 bots / SWARM 2.0 (F2)(Elite)-10.03.07

« on: March 10, 2007, 01:00:18 PM »

Code: [Select]

' SWARM
' (Updated for v2.43)

' Delete birthtie and move

cond
start
.tie 1 *.robage sub 0 floor mult inc
.deltie inc
*.maxvel *.vel sub .up store
stop

' Set eye5

cond
*.robage 0 =
start
1220 .eye5width store
stop

' Swarm

cond
*.robage 5 mod 0 =
*.eye5 0 >
*.refeye *.myeye =
start
*.refaim .setaim *.robage sgn mult store
stop

' Eat food

cond
*.eye5 0 >
*.refeye *.myeye !=
start
*.refvelup 50 add .up store
*.refxpos *.refypos angle .setaim *.robage sgn mult store
stop

cond
*.eye5 50 >
*.refeye *.myeye !=
start
*.refvelup .up store
*.refxpos *.refypos angle .setaim *.robage sgn mult store
stop

cond
*.eye6 34 >
*.refeye *.myeye !=
start
-6 .shoot store
16 .shootval store
stop

' Reproduce

cond
*.body 700 >
start
30 .repro store
314 rnd .aimdx store
stop

' Feed on and store body

cond
*.nrg 2000 >
start
100 .strbody store
stop

cond
*.nrg 500 <
start
*.body 0 floor .fdbody *.body sgn mult store
stop

' Anti-virus (just in case)

cond
*.mkvirus 0 !=
start
*.mkvirus .delgene store
stop

end

35

Off Topic / Invision help - how to limit spammers

« on: March 08, 2007, 04:03:07 PM »

How's this for a CAPTCHA?

Essentially, a bunch of pictures of cats and dogs, and you have to got through identifying them as either cats or dogs. Humans can do that easily, but computers completely fail at that kind of pattern-recognition.

36

Interesting behaviour bots / Anticanni (IB)(Elite)-02.03.07

« on: March 02, 2007, 12:49:39 PM »

A quick sim shows that it breaks quite quickly since the probability of the anticanni mechanism breaking far exceeds the probability of an actual cannibot arising. There's not enough need for it.

I thought of using a bot that randomly converted to canni a few percent of the time, but since cannibotism is disadvantageous in an environment of anticannis I'd expect the cannibot-inducing gene to break.

If poison affected conspecs then that could be used rather than the clumsy counterattack gene, but after modifying the code to do just that a while back I found that poison was too unforgiving - a stray shot could brush an anticanni and it would poison the 'attacker'. Then the resulting melee would result in more bots getting tagged, and it got messy pretty quickly.

A tiefeeder might have better luck, since a bot can instantaneously and precisely leech from the offending tie. Perhaps a second population of hostile tiefeeders could keep the leech gene unbroken, and then cannibots would be disadvantageous since their conspecs would simply be far too defended. Might even see the two populations evolving measures to prevent them attacking each other.

37

Interesting behaviour bots / Anticanni (IB)(Elite)-02.03.07

« on: March 02, 2007, 02:53:58 AM »

' Anticanni

' The first three genes are the anti-canni ones, and then the rest is the bot behavior. Only thing I'm not satisfied with is that that third gene is rather fragile:

def loc 971

' Epigenetic ID

cond
*.robage 0 =
start
.loc .loc store
.delgene inc
stop

cond
*.robage 0 =
start
.loc .memloc store
stop

' Eeek, a cannibot! Fire to blank his ID and mark him as an enemy

cond
*.nrg 1000 <
*.shflav 0 !=
*.shflav -2 !=
start
*.shang .aimshoot store
*.loc .shoot store
0 .shootval store
0 .shflav store
stop

' See food

cond
*.eye5 0 >
*.memval .loc !=
start
*.refveldx .dx store
*.refvelup 40 add .up store
stop

' Shoot food

cond
*.eye5 50 >
*.memval .loc !=
start
-1 .shoot store
*.refvelup .up store
*.refxpos *.refypos angle .setaim store
stop

' Move around

cond
*.eye5 0 =
start
*.maxvel *.vel 5 add sub 0 floor .up store
*.eye9 *.eye1 sub .aimdx store
stop

' Avoid conspecs

cond
*.eye5 0 >
*.memval .loc =
start
314 rnd .aimdx store
stop

' Reproduce

cond
*.nrg 6000 >
start
50 .repro store
stop

end

38

Off Topic / Question about the Feasability of a Neural Network Set-up

« on: March 02, 2007, 02:44:28 AM »

Ah, OK, thanks Num

39

Off Topic / Question about the Feasability of a Neural Network Set-up

« on: March 01, 2007, 11:14:27 AM »

Are any of the programmers here familiar with the workings of neural networks?

If so, have you any idea of how much difficulty would be involved in setting up this system:

Some of the neural network's outputs cause a change in the actual set-up of the network, so it can effectively re-write it's connection archetecture.

The network observes a human programmer making improvements to the network, and learns to make basic improvements to itself. A node is periodically added to the network, and the re-optimisation process begins again (whihc would of course aid subsequent attempts), with the programmer giving the network both an initial push and subsequent reinforcement.

It seems to me neural nets lend themselves to this kind of recursive self-improvement because of their simplicity, modularity, adaptivity and nonlinearity

Firstly, how easy/difficult would this kind of self-improving neural net set-up be to program?
Secondly, to you think a network could learn to improve itself in such a way?

40

Off Topic / A couple of evo stories

« on: February 27, 2007, 11:07:42 AM »

There's a higher level of relatedness between colony members for ants also. In the robot swarm description a similar high level of relatedness occured between the co-operating bots.

An 'us' vs 'them' setup where you have closely related entities in a group, and several groups of entities further away genetically, in inter-group competition, seems to be a good bet for an 'altruism-friendly' environment - if a canni evolves, the tribe is compromised and gets usurped by a non-canni tribe.

So mimicing this in DB would prehaps be several colonies of dependant (on others in the colony) bots with epigenetic conspec recognition (which allows for colony reproduction), competing for limited resources.

A 'simple' and less genetically fragile antbot of some sort might give interesting results when evolved, where there's another meta-level above the level of a bot - in this case a colony.

41

Bot Challenges / Challenge #2: The Maze

« on: December 25, 2006, 02:02:14 PM »

Simulation file not attached

Version required: 2.42.9r and greater

Rules:
Your bot will start in a coridoor on the left. Reproducing is not allowed. Walls are visible to bots. Your bot must have a gene that imobilises it it at age 10,000. There is a cyan bot in an area at the bottom right

Bronze objective: Find your way through the maze to the cyan bot, and kill it with shots, within 10,000 cycles

Silver objective: Find your way to the cyan bot, and bring it back to the beginning of the maze, within 10,000 cycles

Gold objective: Find the way to the cyan bot, then bring it back to the beginning of the maze, taking an optimal path (ie. no trial-and-error, the path must be direct, and you can't cross your own path), all within 10,000 cycles

42

Bot Challenges / Challenge #1: The Ravine

« on: December 25, 2006, 06:16:49 AM »

A saved file is attached. Load it into 2.24.9d and watch the bot on the left. It ties and fixes to make a line, and then shoots out over the ravine like a harpoon or a cobra. As soon as it's grabbed the cyan bot, it contracts to pull the cyan bot back to the left cliff.

This challenge is officially closed. You won't get an award for completing it, although if you want to try simply as a test of your programming skills you are free to do so

I'll upload the new challenge later today.

43

Off Topic / Merry Christmas!

« on: December 25, 2006, 06:11:31 AM »

Merry Christmas everyone!  

And a Happy New Year

44

Bot Challenges / Challenge #1: The Ravine

« on: December 22, 2006, 04:35:27 PM »

OK, I'm closing this challenge on Christmas day. If you've got a bot that beats the challenge, post it, if not, maybe you'll have better luck with the next challenge

I'll post up the bot I made for this challenge then, which ends up with a silver. Then I've got a shiny brand new challenge planned

Hint: It's a maze

45

Off Topic / Panspermia

« on: December 22, 2006, 01:24:20 PM »

Hey, this is interesting ...

Just came across a paper on lithopanspermia in star clusters. Due to the closer distances, it's much more likely that life can spread from one system to another. In longer-lived clusters, life originating on one planet could possibly seed itself to a majority of the other systems in the cluster through the process.

Lithopanspermia in Star Forming Clusters