Harnessing evolutionary creativity: evolving soft-bodied animats in simulated physical environments

[@Jabrils]

this is quite possibly one of the greatest things I've ever seen.

[@polarbear6197]

Yes

[@dbneptune]

I didn't expect to see you here! And it is really cool.

[@Doctorfudge62]

I'm still waiting for that part 2.

[@discountcoconut1458]

How do you have 50 subs

[@Doctorfudge62]

@@discountcoconut1458 I made shitty gaming videos in high school.

[@maxwell_edison]

same where it be

[@Rafale25]

Sexually Active Narwhal His computer is still calculating

[@Hailexx]

Part 2 where u at tho

[@technoeevee6969]

Perhaps a new type of cell that is specifically able to be ridgid would be cool to see what sort of skeletal structures would evolve

[@discountcoconut1458]

Bones no shells.

[@shanedancer3895]

How would joints work? I’d imagine there would need to be another special cell for that but it could become too complex and wouldn’t work

[@technoeevee6969]

it might work the same way joints work for us- a gap between the two seperate bones to allow them to bend. They won't have cartilage holding them together, but fortunately these simulations won't require it ^^

[@shanedancer3895]

@@technoeevee6969 The problem is I think adding joints and bones begins to make the simulation too complex and could result in a situation where everything with a bone and joint just ends up losing because they take too long to become better than the simpler other versions

[@technoeevee6969]

@@shanedancer3895 *Shrug* still worth a try, right?

[@trudyneo]

Love the jelly look of the creatures. Can't wait for the code!

[@OminousPinapple]

I would love to see a game based around this! Maybe a "God-like" game where you first come up with a shape, give it an environment, (ie,, water, ground, food source) you drop it somewhere, and let it go to see if it can evolve to survive. Keep up the great work man!

[@johnsmith-sp6yl]

there are a few games like that. can't remember the names of them, but the randomly generated life forms were creepy as all shit

[@nedrynthecheeto66]

OminousPineapple I've heard of this game called "Species." I don't know where to buy it, but I've seen people playing it on KZbin. I think that closely matches what you're looking for.

[@cheetoschrist5685]

Species is free to play. just search for their website and you can download it.

[@thomashewitt8104]

I played a game at EGX in the UK a couple months ago called “Ecosystem” - it’s exactly what you’re describing

[@somewhatsour]

Spore

[@karlmuster263]

The fish example at the end is interesting. You'd think the spine comes from the biology of animal development, transporting nutrients, protection, etc. But it's functional just based on locomotion. Maybe bony fish evolved just to get away from predators or catch up with prey.

[@linsproul3548]

the spine wast actually evolved for locomotion in this simulation. it was just a section along the middle of the fish that would gain no benefit for being muscle cells. just think about it, the fishes fins wouldn't move left or right if it expanded or contracted cells in the center, so there was no point for them to be there

[@thomasrosebrough9062]

The spine probably has all those uses and more. It's a pretty integral part of our biology.

[@robokill387]

@@linsproul3548 true, but that's probably why the spine is in that specific location.

[@Sara3346]

How about eveloving them with two goals, 1. To move. 2. To collect reasorces. ?

[@VoltzLiveYT]

That can be reduced to one goal: collect resources. You have to move to collect resources, so it would evolve to move.

[@JNCressey]

+VoltzLiveYT, Not necessarily if you're in water and the resources can move to you. :D

[@VoltzLiveYT]

JNCressey but a creature that can move, it will have access to more food, and thus be more successful than one who doesn't

[@JNCressey]

VoltzLiveYT, depends how much the ability to move ends up costing it resourcewise as to whether it's worth moving.

[@maxwellsimon4538]

You would get different sets of creatures that would collect resources in different ways, whatever ends up evolving. Some may move to collect, others may sit and wait, and others still would move resources to them.

[@AmritZoad]

0:06 When he said "Animats". I felt that.

[@Table53]

Three year old video, claims there'll be a part two at the end... I'm still hopeful.

[@Gunth0r]

Half Life 3 confirmed

[@revimfadli4666]

@@Gunth0r half life 3/2 is out tho

[@DrWrapperband]

Super work, please release to Github, a lot of people are starting to get interested in evolving systems and this is great stuff.

[@PhilippeLarcher]

Anyway after a few billion years it will evolve into creationists

[@ryanmapping7944]

Some ways to add realism would be an digestory system to eat food, energy costs and many creatures in one ecossystem, also 3d would be cool, althrough we would probably not have particle physics for water then without serious optimization but the digestory system idea would be more realistic with it.

[@AGoldSoldier]

Seems the algorithm decided to show this video to a quite a few new people. Myself included

[@jd233joel4]

A Gold Soldier me too

[@PhilippeLarcher]

It's evolving too

[@Joona1410]

When do you think it'll be avaible for download?

[@mjoach]

Working on it whenever I get a chance, but with the current speed, it will probably take me another month.

[@Joona1410]

Michał Joachimczak Okay. I'm really looking forward to it. No need to hurry, though.

[@mono4171]

this is so cool dude

[@blorger]

omfg please release a beta

[@philipbraatz1948]

A month passed, I wonder how its going.

[@sapiranimations]

OHMYGOD I saw the videos you did 3 years ago, I was so excited about them. and now you do this!! and something even cooler soon!! that's so awesome

[@mjoach]

wow, thanks for that, glad you liked it!

[@No-oneInParticular]

It's been over 2 years now... can I see the next video? Please?

[@comedymask5340]

This is a such a cool simulation. I love how they can change to their enviorment.

[@PaleyDaley]

Wow, this is truly the best ALife example I have seen since Karl Sims.

[@Sty1a]

Every week i check this tab i left open... i cant wait for the source :D

[@ppraisethesun]

PM me when it's here

[@MiniDemonic]

I don't think he's checking his tab every week. Source is never coming and the particle liquid video is never coming.

[@comradecameron3726]

Rec0iL you still checking?

[@LimeRavioli]

Is it here, yet?

[@NinjarioPicmin]

@@LimeRavioli i dont think so

[@sammyblaze4234]

2 years later, no follow up... :(

[@Moon-ge8mc]

Sammy Blaze he's a stupid liar

[@thetaai8262]

i feel like something like this could be used to see what creatures on other planets may look like

[@dontknowdontcare1934]

Yeah that would be cool

[@Comfie]

That would be really cool

[@PhilippeLarcher]

Exactly

[@shadowfire04]

that is so cool to see how evolution came up with all these solutions by itself, and how we can see how these solutions developed.

[@pootimcbooti]

these are so cool! i cant imagine how other lifeforms on other planets look like

[@conradmader2264]

You think that's hard? I'm doing an entire project on scientifically plausible aliens. Sorry if you got the impression I was bragging, I wasn't.

[@tamsynmaginness3282]

Please post again!! I love this video so so much, it is so cool, and I would love to see more of this!! And more videos, this got me really interested in Evolution. Please post again.

[@MultiSciGeek]

This is so simple but so complex at the same time. Can't wait to see what else we can come up with.

[@pikminfan6778]

1:07 No concept of leg or tail, eh? Well, what's that flexible whippy thing on the creature's end? And what are those two stubby things it's running on?

[@jamesgabor9284]

You probably aren’t reading this but I really want some more content like this. I think it’s really interesting

[@toxicreaper4632]

@Michał Joachimczack is this available to the public, because I'm working on a bachelors in radiology, and majoring in evolutional theory, and this would be amazing for my spare time as a game to help me understand muscle flow.

[@dark4king]

This is just amazing. Neural Networks are the closest way to understand nature. Wait, when I see the fish "developed" a spine, we already understanded how this world was made right!

[@4.0.4]

Mateusak These were not neural networks but genetic algorithms. Neural networks are more about developing a general process than mutating something into solving one task

[@mjoach]

Actually there are neural networks there, this is what controls behavior of each cell :) In that context, we call them artificial gene regulatory networks, but computationally, they are the same thing.

[@cicadafun]

Michał Joachimczak Is it possible to have cells differentiate for different bodily tasks?

[@revimfadli4666]

@@mjoach so are you using something that works like a CPPN to control the growth of cells that follow simple signal propagation rules, or are the cells themself neural network-controlled?

[@Zackapo]

You should make the animat loops longer, some of the animations were too fast to register properly. great job!

[@SerAlbi]

I rarely comment on youtube, but this is mindblowing! 😵 Please tell me this is open source or at least if there's like a standalone version that we can experiment with! 😍

[@NICK....]

I have two questions: A: How much computing power is required to run them; B: Can I, a normal user, download the programs required to run them;

[@henrycgs]

I need this so much in my life! I need to put this on for weeks, so when I come back I see a fully walking huge animal

[@AaronEstesJ]

I have been waiting for a new, well made genetic algorithm video for a while. Thanks! If there are any more similar to this, I would love to see them!

[@weak1ings]

soon! Thanks for working on downloads!

[@MrElrolis]

This would be pretty cool if used for a 3D model, I wonder if there is a program like that? I've seen about motor evolution but not physical.

[@jasperlusson5797]

wheres part 2

[@XregularC_Casual]

I'm learning to program in school and I love it. I'm not nearly capable of creating amazing things like this and it's making me quite jealous. XD

[@jakelodwick]

XregularC Just keep at it and you’ll be at this level before you know it

[@bragunetzki]

Ahh, I still remember making sure to check out when the hew part comes out...

[@phlimy]

This is awesome! I can't wait for the even-better-how-is-it-even-possible thing you previewed, so hyped :D

[@MultiSciGeek]

Where can I learn more about this? Is there any talk or presentation about such projects?

[@newM0nkey]

So cool! What physics engine? Any source code you can share?

[@mjoach]

Most of the exps in the video were done with my soft-body physics implemented on top of rigid-body part of the Bullet Physics engine. I'm in the process of replacing it with a much more stable and much faster engine and I will open source it in a few weeks.

[@HolyGarbage]

Hi, I'm doing similar research/hobby projects and so far I've been experimenting with Bullet/Box2D using their existing framework to implement soft bodies but this proved to be really computationally expensive to do properly. I am also not very educated in physics so a ready made soft-body physics engine would really stream line my projects to get to the good stuff and not have to worry about the physics. Is this engine of yours up yet and is it available for download? Do you got a github account or similar? P.S. Also want to say that I'm a huge fan of your work, this kind of stuff is what I'm striving to accomplish myself! Cheers from an undergrad at the University of Gothenburg, Sweden! P.P.S. Does your engine also include the fluid mechanics and particle diffusion (morphogen) seen in many of your videos?

[@mjoach]

Hey, thanks a bunch for the kind words! The code is not on GitHub yet, it is still a terrible mess after I decided to replace Bullet Physics with LiquidFun (check it out!), but I am working to bring it towards more general usability right now, so hopefully people like you can find it useful :) As for fluid mechanics as you can imagine it becomes very costly if we try to combine it with evolution, but I am indeed experimenting with it (hence the use of LiquidFun). And as for morphogen diffusion I do not explicitly simulate it, I only use a very simplified approach to provide some basic means of direct communication between cells. I tried more realistic approaches in the past, but couldn't justify the added costs. Anyway, feel free to get in touch if you need some pointers!

[@HolyGarbage]

Ah yes, very costly indeed. My projects are normally open continuous simulations which means I'd like a couple thousand (at least) organisms simulated at the same time rather than having an explicit fitness function. I once experimented with liquid dynamics and diffusion as I wanted to simulate concentrations of various fluids such as pheromones, poison, resources they need to proliferate etc. But simulating that over the whole world space with an acceptable resolution proved way to costly, like several orders of magnitude of an acceptable level... I have experimented with gauss/normal distribution around discrete "macro" particles that represent the field's local centers of mass as of late which seems to be extremely more efficient. Do you use something similar? Also, something I've found very difficult, how do you calculate the forces of your swimmers? I mean a surface moving and pushing is pretty trivial to calculate the forces but things like pressure I have no idea how to simulate without a discrete density grid liquid particles. It does look like your octopus looking creatures utilize pressure when they swim, do they and how did you solve that in that case? Thanks, I'll check out LiquidFun, thanks for the tip!

[@mjoach]

Well, it's really hard to make open worlds and my feeling is that usually the complexity of what evolves there is lower than what can be achieved with some simpler evolutionary algorithm (if complexity is something one cares about...). But it's largely because of overwhelming computational costs and the fact that it is so much harder to control the direction of evolution in an open system. One needs to tune things here and there to prevent trivial local minima and that takes time. So good trade-off between realism and speed is the most essential thing here. As for diffusion gradients, yes, I've been using normal distributions around macro particles in the past as well, so I think it is a reasonable approach if one cares about speed. As for pressure forces - not sure what you mean. In this system, there is no explicit concept of fluid pressure. If you mean fluid drag, there is just one simple equation that depends on the angle of attack of an edge against "fluid".

[@Darfail]

VERY COOL! what software allows you to have soft hexagons like that?

[@mad_machine4850]

this is fantastic! I'm curious to see how far and complex it can become

[@seeme00]

Cool work :) You mention "A rudimentary model of fluid drag". Do you have any pointer on that? Do you use a third party middleware for the physic simulation (Box2D? Bullets...)? Thanks!

[@mjoach]

Yes, most stuff in the video was Bullet Physics, thought it was it's rigid-body part bend by me to simulate soft-bodies, not the best way to do it. I now rewrote it to use LiquidFun (a mod of Box2D), which seems much faster in these scenarios. As for fluid drag, check the simple equations in our older GECCO paper: goo.gl/Yt2UKt and you can find original references there as well.

[@revimfadli4666]

I really like how this used graph network before it was cool

[@mischake]

I live this to bits. It's exactly what we should be working on, so away with hollow avatar systems and actually put biology in things like games and simulations

[@fauna575]

This is honestly so fascinating and amazing! As a bio student, this is making me incredibly happy

[@ZtereoHYPE]

Hmmm... where is the continuing? I want to see

[@herbexionreviews2039]

still waiting for part

[@SangerZonvolt]

Looks very nice. Are there any updates on this?

[@Krishna-f5r8w]

Is this project still active? It's immensely interesting.

[@huglebunnys]

Hey I'm really interested in the state of computer-programmed evolutionary algorithms and projects, do you use a site for recent information on it or a forum? It's hard to find current videos on youtube at least but this video is only 3 months old compared to most of the others that people upload, which can be up to 9 years old and a bit dated.

[@mjoach]

I don't run a website, but maybe check out the proceedings of Artificial Life conference, where many researchers from this field meet and show their work. You will always find a few papers with some artificial creatures there. The proceedings are open access: mitpress.mit.edu/books/proceedings-artificial-life-conference-2016

[@rebelarmstrong4310]

Where do I find this? I really want to use it.

[@dreamer097]

Can you tell me about the particle fluid at the end there? I'm interested in the reaction diffusion patterns it makes when it settles.

[@mjoach]

Those fluid particles are only subjected to elastic forces, so I suspects patterns are either Moire-like rendering artifacts or a side effect of how physics is implemented (e.g. how neighbors are selected for each particle at a given time step). Didn't investigate it further though.

[@dreamer097]

thanks, I was suspecting the same thing. But I was hoping you could tell me what program or code was used for it. It's not Algodo is it?

[@mjoach]

It's Google's LiquidFun.

[@PasseScience]

Wow, that one was really beautiful. What's the 2021 status of it?

[@Moon-ge8mc]

he lied

[@PasseScience]

@@Moon-ge8mc What do you mean?

[@Thekeygen36]

Man i love this kind of work! Keep it up! Cheers from Argentina!

[@TheDopeyElephant]

is there any website where i can make a creature then see it evolve with future generations?

[@explosu]

Did you ever consider/try adding apoptosis? Would be interesting to see a system like this work out movement using only a regulatory system for growth

[@fergheinman394]

so whens part 2 gonna come? we're only 2 months away from this being 3 years old

[@BuddhasRandomTravels]

is this project ongoing? any source files?

[@Krejzifik]

Kiedy to wyjdzie?/When will it release?

[@gyurto]

When will this be ready for download?

[@hmeric7495]

when will it come to a github near me

[@aliciabaumgartner1406]

Fascinating! Will you consider making a video explaining the gene regulatory network? I perused your research paper but i'm having difficulty understanding the process.

[@galesx95]

The part it says "every cell is controlled by evolved gene regulatory network". What does it mean? I'm having a hard time wrapping my head around that. Thanks!

[@mjoach]

well, the point here is to mimic some of the not very intuitive solutions that biology came up with. In the case of multicellular organisms that would be having a copy of something that has all the necessary information to build the whole organism (i.e., DNA) in its every cell. Seems wasteful :) To the point: DNA encodes proteins that make us, but more importantly also the logic of when to produce them and in what quantities (e.g. if gene A is active, and gene B is not active, than activate gene C). When biologists try to understand what DNA of an organism does, they represent this logic as a gene regulatory network (GRN). So what we do here is we evolve a virtual DNA that encodes virtual GRN. When virtual embryo grows, a copy of this virtual DNA (and a GRN) is placed in every cell and with this GRN each cell _independently_ determines what to do at a given moment (such as whether to divide or die).

[@youngjunlee1108]

Great work, may I ask you what engine or environment for making this project? Thanks.

[@nethacker91]

He answered elsewhere saying it's the Bullet Physics engine.

[@gamez1]

3:19 "coming soon" what do you mean by "soon"...?

[@PhysicsGuru100]

Fantastic video!

[@gaboderflinger7854]

This is so interesting, why you stopped making this type of experiments?

[@zeektm1762]

Im interested in running simulations like this but I have yet to find one that I don't need knowledge of coding in. Is the source avalible in an easy .exE?

[@GraveUypo]

what about rigid cells? what if you give them the ability to grow bones?

[@peterganunis5824]

Any updates on the release date? Thanks!

[@lordkermit4657]

Can we get this on our phones and stuff?

[@cariboowho]

Are you going to release the code? I glanced through the description and I didn't see it, though maybe I just missed it.

[@mjoach]

(see reply above)

[@mewt5358]

you think we could use this ourselves? I really wanna play around with this program

[@alexey_guzey]

Hi! Do you know when the code will be available on github yet?

[@KordaroKnow]

Imaginations and simulations drive evolution throughout the nation... 😯

[@bapanada9446]

where must I study to learn this kind of stuff?

Hi I see you haven't shared any video after this. Can you share your paper with me?

[@teawrecks1243]

1:19 ah yes, the crippled seal. truly a marvel of evolution

[@lookolookthefox]

so where can we find the program?

[@davidmorrill1230]

what is this and where do I do find it?

[@sapiranimations]

did that coming soon ever came?

[@Skythedragon]

Next step? 3d of course!

[@junosoft]

Is it ready already?

[@olekrarup9570]

How about "magnetic" creatures? Let a magnetic field point into or out of the screen and calculate the force on each cell based on F=q(B x v).

[@mjoach]

But why?

[@olekrarup9570]

+Michał Joachimczak I thought it would be interesting to see how evolution would tackle an enviornment with rules that are completely different from Earth's.

[@mjoach]

fair enough

[@maxwellsimon4538]

Are all of the animats given the opportunity to grow rigid "bones" and that just turns out to be disadvantageous in most cases? oO is that option only available for optimizing the movement of a fixed shape?

[@mjoach]

No truly rigid parts in these exps, but they are easy to add in the new engine, so I am planning on doing that. As for passive soft tissue, we also did that when shape was evolved, check the end of the first 2014 paper linked above for some example. In general, if there is no cost to having and using muscles, evolution tends to evolve creatures that are 100% muscle. It just allows them produce the most power and convert that into motion. So we have to always add some cost for muscles if we want evolution to allocate them in a meaningful way.

[@maxwellsimon4538]

Michał Joachimczak Ah, that makes sense. Having cells that don't contribute to movement, are as good as dead weight in these cases. I can't wait to see future videos from this channel!

[@dividedreality9708]

I can watch this for hours.

[@princess-celestia]

I love this. I stand in awe of your coding prowess. I see you tested terrestrial and aquatic life. Any attempts at avian, flight-capable life? Perhaps allowing the animats to convert some cells to lighter, but non-actuateable, solid cells to facilitate forming "feathers" for wings? Or maybe convert muscle tissue to lighter-than-air helium bladders? Out of curiosity, have you tried incentivizing the animats to minimize fuel consumption (amount of contraction)? How long (in terms of hours [or days?] spent running the program) did it take to generate the creatures in this video? Once the code is available I'd love to try it for myself, but I'm afraid it'd make my computer explode.

[@conradmader2264]

When I first saw this giant block of text I thought it was going to be a giant essay on how evolution doesn't exist. Thank you for being a sensible human being.

[@sorgan7136]

is the program ready?

[@pabloyammix]

thats actually amazing! the spine epiphany even. Waiting for part 2 also..

[@planespottermerijn]

How is the code comming along?

[@eliotjanvier4653]

Hey can we download that? :D

[@sploopst6868]

that fish with the active flanks only is fascinating

[@purukumi4957]

Whete can I download this?

[@romanrhodes423]

this video came out about 3 months ago. do you have a release date planned?

[@romanrhodes423]

Also, can you make a flying creature by just disabling gravity?

[@ozvoid1245]

In real life, flying needs gravity. Creatures fly using the ability to glide, and using wind and air. Basically, he would have to add gliding cells, and wind.

[@_DarkEmperor]

Guhuru, widzę że stałeś się sławnym KZbinrem :-) 150tyś odsłon.

[@guysimple8491]

If the task will be "to create an optimal behaviour program for surviving in african savan for upright hominid" will it create consciousness?

[@TheYeetedMeat]

Depends. If anything, it might go straight for a herbivorous altruistic route.

[@whippet_boy8594]

If this simulation came up with a rudimentary tail, legs and spine without any prior knowledge of evolution or life, then maybe this could tell us that life on other planets maybe is not that different and it follows a basic structure?

[@3gunslingers]

Life on an other planet will be very similar in the basic layout. Yes.

[@LineOfThy]

yes. yes that's exactly it.

[@revimfadli4666]

"coming soon!" 7 years later: