Scriptbots v4: Open Source Evolutionary Artificial Life Simulation of Predator-Prey dynamics

Подписаться 3,4 тыс.

Просмотров 83 тыс.

50% 1

My continued work on evolving crazy hunting communicating things! :)
- Bots can specialize in processing plant food or meat, and can therefore become carnivores or herbivores.
- Bots can hunt each other. They can communicate by flashing colors and also shouting at each other.
- Bots can choose to share food with each other. (allows for altruism)
- Both sexual and asexual reproduction is implemented.
- When a bot dies, food is distributed evenly to bots around the event. This leads to emergence of both scavengers and hunting packs.
Improvements in version 4:
- New brain (Damped Weighted Recurrent AND/OR network)
- No herbivore/carnivore dichotomy. Food specialization variable instead. Allows omnivores.
- More sensors were added (clock sensors, blood sensor, hearing sensor)
- More actuators added (ability to share food): can lead to altruism
Website:
sites.google.c...

Опубликовано:

1 окт 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 96

@dialupdavid 9 лет назад

Honestly incredible, it just blows my mind how once you create a program that can essentially thinks for itself, it goes off to adapt its own code to best suit the simulation.

@oisiaa 2 года назад

Andrej is now arguably the world's leading expert on artificial intelligence. He leads AI for Tesla.

@richm6633 8 лет назад

It would be interesting if you could add geographical barriers that slowly formed and slowly dissolved at random, which might lead to examples of speciation, founder effect, bottle neck, etc.

@wolframstahl1263 8 лет назад

That is pretty much what I was just about to suggest =)

@BlueGlowingLight4 8 лет назад

This! Even if it was only duplicating the existing creatures into a separate simulation for 30 minutes then putting them back together.

@HermanWillems 8 лет назад

I agree, because "weather" and surrounding area's create diffirent species.

@wesleydickens9283 9 лет назад

Here we are 7 years later. How many generations of this simulation have lived and died? Have they developed tribes and communities? Has homosexuality developed and prospered in the simulation? Are they making tools? Building shelters? Cities? What predators are they having to fight against for survival?

@lorenzvo5284 9 лет назад

Wesley Dickens Have they gone to Space yet?

@kaidatong1704 9 лет назад

+Deer Viehch Spore reference?

@lorenzvo5284 9 лет назад

Kaida Tong maybe

@calebshade4991 8 лет назад

That would be interesting if he started a simulation with sexual organisms and they evolved to be asexual. But in this current simulation there would be no point to homosexuality as the organisms can't feel love, pleasure, or would get any reproduction pros from being homosexual.

@EpicFishStudio 7 лет назад

do they believe that someone created them? lol, okay, okay, i'm leaving.

@graczmisiek 9 лет назад

It seems like primitive forms of life like to dance around fire

@DustySquitoNM 9 лет назад

This is legitimately impressive work, man.

@arcadegamesify 8 лет назад

Nice. You're one of the people who got me interested in this. I'm currently coding a simulation of evolution that uses neural networks (similar to NEAT), and I'm gonna post a video online about it. Think I'm gonna make several, cuz I have lots of ideas.

@revimfadli4666 7 лет назад

Good luck with your project, cuz I'm looking forward to it!^^

@KarlPhillipBuhr 10 лет назад

This is the most scary description of a working artificial life simulation I've ever seen, good job!

@drgutai1 6 лет назад

Just discovered the guy who made this is now the director of AI at Tesla! I had a great time playing around with this software for hours back in 2011 because of how original it was. Surprised and really glad to see how far he's gone.

@hahathatsgood 5 лет назад

drgutai1 I was surprised when I found that out too! I thought this channel was just some random internet person but he turned out to be one of the leaders of the whole field of AI

@droopsmoop 5 лет назад

@@hahathatsgood You know what this means... HE'S TRYING TO CREATE AN EVOLVING ECOSYSTEM MADE ENTIRELY OF TESLA CARS!

@pauljohnson6806 9 лет назад

impressive.. my first thought is why do the plants not have life and the ability to respond/ defend themselves..? this would make the herbivore/ carnivore choice more evenly weighted.. also.. have you done anything about different life cycle/ breeding length and parental investment weights..? i.e. variable rates of reproduction and many little investment offspring versus few large investment..? all very fascinating..

@revimfadli4666 5 лет назад

If I'm not mistaken, there's another one of this kind of simulation that enables you to even make a custom food web

@Т1000-м1и 3 года назад

96th, just realized, the color stuff is like evolv.io (even tho evolvio came after, but still)

@35906 9 лет назад

I was wondering about a variable or "gene" for sexuality. If they are able to reproduce with only similar objects, or completely different ones. That should be a random variable. Then run it for a long long time, and see what seems to be most common, the ability to reproduce with anything, or the ability to only reproduce with similar "species."

@NuevoVR 7 лет назад

now we're in 2017, recreate it in a 3d environment. also give yourself an immortal avatar and interact with them through it and see what happens. they could end up worshipping you and forming religions based on you. you might end up creating AGI. maybe we need quantum technology to achieve this but its an idea lol

@ConsciusVeritasVids 2 года назад

I tried to intervene even in the early stages by controlling one of the creatures and it died immediately after moving a short distance. Just goes to show you the Programming Gods should not interfere with the creation.

@badmephisto 13 лет назад

@fu11fu yeah but like i said, it doesnt actually matter. Any nontrivial dynamical system will do. Basically.

@somezw 5 лет назад

This is great! I regret not having found this channel before! Simply AWESOME....👍👍👍

@computercat8694 8 лет назад

1:34 Herbivore joins the hunting ring

@Zwerggoldhamster 8 лет назад

So the bots can change color at will? I am surprised that they wouldn't do that more often. Does that mean, that it's beneficial for both carnivores and herbivores to who their kind is?

@revimfadli4666 5 лет назад

Sometimes they differentiate between themselves and prey/predator through color. Then again, sometimes predators and prey mate, resulting in predators that move & look like prey, backstabbing its prey siblings once in a while, or prey that hides among predators

@Vincent-kl9jy Год назад

I am a PhD student who studies evolutionary dynamics and the evolution of intelligent behavior. If you are ever interested in talking with me, it would be a pleasure. I was inspired by these videos back around 2012, and it is some small part of how I ended up where I am today.

@phildarcy9988 7 лет назад

Thanks for sharing this project! It is one of those projects I discovered years ago which really inspired me to learn to program so that I could create my own artificial life simulation, and I did :)

@luccahuguet 2 года назад

Woooow this is NEAT I feel like I have wasted potential and should up my AI game, thanks for the inspiration

@Checray 12 лет назад

I ran my sim for a few hours and had a balanced predator/prey ratio. But the prey started to fight back in an interesting way. They would get behind one another so that they are just out of reach from "spiking" one another. They would make this conga line/train into a circle so that no predators could actually get behind them without getting killed. this basically wiped out the predator population so that there are no more "evolved" predators left.

@shad0wr4ven 12 лет назад

I know you've heard it a hundred times or more, but your simulation is just amazing. By far the best I've seen... I also believe there is a lot of potential. Have you considered any way of allowing occasional random genetic mutations to alter the actual physical capabilities of the bots in relation to each other and their environment? (IE: having larger spike, more than one spike, becoming larger, smaller, slower, or faster, etc) I'm going to try to build a 3-d version which uses your code.

@grox2417 5 лет назад

Anyone from 2019?

@BlackProductionInc 11 лет назад

I wouldn't recommend starting to code in C++. Although it has very interesting applications it's not very easy to learn programming concepts because you are often very close to the hardware. Once you learn to code properly you can turn to C++ prettty easily because the basic structure is the same. But you rarely really need the low-level capabilities of C++ these days. Search on google for 'codecademy' they offer a really good inter-active learning environment for free.

@Isaac_howell 12 лет назад

well this is just bizarre,I have a stable environment with 400 carnivores and 200 herbivores, the carnivores do not seem to move, but form large clumps and simply wait for the herbivores to drift over them, the herbivores drift from food source to food source, this behavior started about 30 mins ago with the program speeded up.

@thethingthatwants 11 лет назад

One thing i would love to see would be this "brain evolution sim" and then mix it with a "morphology evolution sim" so that how the neurons connect can evolve alongside body parts, even seeing how new neurons would connect to new body parts. No idea how it would be done but I would still love to see it.

@TrollyTrollson 12 лет назад

i've been playing around with this for a couple hours now and the evolution seems to mostly be behavioral... ever thought of allowing changes to the eyes??? i think that it would add ~so~ many more behaviors if the eyes were allowed to change even one degree in any way from thier current setup =)

@dirtpoorchris 7 лет назад

thats one amazing simulation.... The fuckin future... Wow man..

@TrollyTrollson 12 лет назад

also i'd like to add that my herbivores developed an interesting method for dealing with carivores, they simply became a part of the circle of death... i even noticed at one point that one of these death circles was led by three herbivores XD

@potato-hj9nm 7 лет назад

This is great but could someone explain to me how to run the program because none of the instructions make any sense to me.

@DrunkGeko 8 лет назад

I ran the simulation myself for about an hour and a strange beheviour became very very popular Big chunks of agents just standing still in a group and some of them turn blue when they see a blue moving predator Predators do not seem to act aggressively at all to those chunks of still agents, which makes sense since they become of the same color What does not makes sense is how those chunks can eat, the food on the ground should not be enough for them to survive and spread (sorry for bad english, i'm italian)

@endorsedbryce 7 лет назад

WOW, that's interesting. perhaps they found a way to survive colonial by sharing all the food that spawns over the area they occupy?

@DrunkGeko 7 лет назад

EndorsedBryce from what i can tell, they survived from a "glitch" in predators and pray's brain: when a predator or a pray got into those big chunks, they started spinning or just standed still, slowly wasting energy and as soon as they died all the members of the big chunk ate the corpose of that stuck agant

@endorsedbryce 7 лет назад

DarkGiona98 weird

@dirtpoorchris 7 лет назад

Just wait till they devolop politics and learn to stay still unless only a few straglers are left for eating. I guess they would need a regional communication + friendly corpse memory function. (in the simulation.... Noone can hear you sound your alarm if your 53 pixels away...)

@Dirtfire 11 лет назад

Correction: simultaneous red and green occur in the javascript version, but not in the C++ version, which I'm currently playing with. Could blue have something to do with scent-emission?

@GameChanger00100 12 лет назад

Circles of death always seem to be moving counter clock wise I wonder why?

@ThaFacka 13 лет назад

wow, astonishing!!! please make it even more complex and add the ability to "ride" on of the agents, in order to graspt whats really going on in one individual!

@RandomInternetDog 13 лет назад

You are so much smarter than me. Been working on getting a evolutionary simulator working for a awhile now and I think I just don't have the knowledge to succeed.

@PasseScience 3 года назад

Great! any 2021 update on this ?

@ConsciusVeritasVids 2 года назад

I believe the original creator no longer maintains or updates the app, but it's been picked up by someone else and now has a Windows installer for the program which now goes by "Evagents"

@ihavetoshutthewindow 12 лет назад

can someone please tell me if you automatically save your simulation or not?

@Aceionz 12 лет назад

Is it possible to speed up this program? So that you'd get to see the results of the evolution more quickly....

@Dirtfire 11 лет назад

I'm not absolutely sure, but the blue marker appears to be the sharing function, or give actuator.

@samuel2437 11 лет назад

What does the epoch refer to? The generation? And how did u calculate the agent's age

@zes3813 7 лет назад

wrg, no such thing as succesx or notx. ts not for funx, not nerx

@revimfadli4666 7 лет назад

Would using long-term memory units(like GRU and LSTM) end up with more interesting behavior/solve problems like the circular pack?

@Misthema 8 лет назад

Neat! How do these bots learn? Is it purely evolutional or is there some kind of RL in the play?

@violet_broregarde 7 лет назад

Yeah they spend their entire lives just doing the thing they do when that thing happens. The ones who do the best things get to have babies that do similar things.

@MisterJimLee 11 лет назад

The next one should be where new neurons can evolve with new functions

@PlesiothxKichigaino 7 лет назад

AMAZING AMAZING ART ART ART! ! Beauitful magic! Haaa!

@rafaelcorrea7959 10 лет назад

Did you stop your work on this? The results were amazing!

@zippolag 12 лет назад

Same question I left in your older video '**

@darkcaster6572 9 лет назад

Can someone help me it says that there is a virus when I open the game should I run it or not?

@treywaka2444 8 лет назад

+Darkcaster65 I doubt he'd post a virus along with the source code. That'd be pretty stupid. Either way you could just follow the instructions and compile the code yourself: github.com/karpathy/scriptsbots

@Markus9705 10 лет назад

Cool.

@Pabce96 12 лет назад

Hi there! i've been following your videos since your second simulation and I must say that it's fantastic. I became so enthusiastic about it that i decided to learn some progrmation my self to do something similar. Right now I have a simple predator-pray evolutionary system, but my main issue is fluidity in the simulation. It runs so slow and jittery its a pain to watch. Do you have any ideas on how i could improove this? Thank you so much and keep on the fantastic work!

@Utheree 8 лет назад

I understand nothing but this looks beautiful !

@DrZale 12 лет назад

I really love your life simulation work! It is very cool that complex behavior such as we see in our world can evolve so quickly with such "simple" creatures. I've started my own simulation project, and although its not nearly as trivial as I would have hoped, one day I hope to create something like this myself!

@Isaac_howell 12 лет назад

All my herbivores formed "rings of death" instead of the carnivores, very interesting. Congrats on the program it's amazing

@MisterJimLee 11 лет назад

You should implement smelling where the agents detect smells (pheromones?) from other agents instead of the movement around them.

@arcadegamesify 9 лет назад

How do I enlarge the simulation?

@Voltroll 11 лет назад

Thank you very much for putting your personal time into this, I've gotten great enjoyment out of it. You have my subscription.

@patrick1020000 13 лет назад

O___O complex. . .

@lookoutforKURT 13 лет назад

I'm scared that the simulation theory actually sounds somewhat possible.

@thepineapplegamer6872 6 лет назад

lookoutforKURT why is that scary

@meyerjac 12 лет назад

They're kinda cute.

@Splatball 13 лет назад

Really well done. Please continue to make these.

@DanielFenandes 9 лет назад

So much sensors

@evanvitvious2075 11 лет назад

You're amazing! How did you learn to code?

@BlackProductionInc 11 лет назад

This my friend, is fucking awesome :D

@cyotokatrovedmond6408 9 лет назад

How do I install?

@bkboggy 10 лет назад

Very impressive!

@Sunoco 13 лет назад

This is so awesome.

@Taumantis 10 лет назад

I just looked at code, so my first expression is I know why i prefer C# ;p and second thought it's kinda "evolution of parameters" and optymalization. Sadly there is not neural network with witch it could be very impressive. Or i am missreading something. BUT STILL AWSOME JOB :)

@Darfail 9 лет назад

previous versions used a multilayer preceptron model which is a "real" neural network but you'll notice here that his rule-based network is recurrent and has internal storage values. that's...pretty AWESOME MAN. and the new network certainly don't perform worse than the old ones.

@Taumantis 9 лет назад

DarwinThug Oh so I have take a bigger look:)

@Darfail 9 лет назад

Taumantis if you do PLEASE contact me again, I'm C++ illiterate but have tons of questions about the project