Playing around with reaction-diffusion_

Tim Hutton - 2012-05-10 15:40:33+0000 - Updated: 2012-05-10 15:40:33+0000

Playing around with reaction-diffusion systems in Ready. This video shows two coupled Gray-Scott systems. The large spots can only replicate when full of small red spots. Likewise the red spots can only grow inside the large spots - outside of the large spots they die out. The large spots usually inherit their small spots when they replicate - those that don't will no longer replicate. In the starting pattern only the central spot is seeded with small spots, so the other two spots don't replicate at all.

Two coupled Gray-Scott systems. The large spots can only replicate when full of small red spots. Likewise the red spots can only grow inside the large spots - outside of the large spots they die out. The large spots usually inherit their small spots when they replicate - those that don't will no longer replicate. In the starting pattern only the central spot is seeded with small spots. The other two spots don't replicate at all. Created with open source software: http://code.google.com/p/reaction-diffusion/

Shared with: Public, Felix Woitzel, Xabier Barandiaran, Tom Froese, Gerd Moe-Behrens, Matthew Egbert, Ezequiel Di Paolo

+1'd by: tagesu abdisa, Gerd Moe-Behrens, Felix Woitzel

Reshared by: Felix Woitzel, Hiroki Sayama

Hiroki Sayama - 2012-05-10 16:06:28+0000

Amazing!! This could be extended as a model of multi-level selection in evolutionary systems.

Tim Hutton - 2012-05-10 19:16:39+0000 - Updated: 2012-05-10 19:21:47+0000

Thanks Hiroki! I was thinking of putting some kind of information-replicator inside, instead of the red spots. Then the big spots would play the role of the lipid membrane and cooperate that way. But I'm open to ideas.

Hiroki Sayama - 2012-05-10 20:25:56+0000

How about making the parameter values of the red spots also dynamical variables and letting them change/diffuse over space? Another suggestion is to make unsuccessful green spots shrink and die, which should be easy to represent in the equations.

Felix Woitzel - 2012-05-27 00:27:04+0000

Laplacian Plane Deformation

Tim Hutton - 2012-05-27 09:48:17+0000

+Felix Woitzel , you mean an area-preserving distortion? What for?

Felix Woitzel - 2012-05-27 10:12:14+0000

I mean like a gradient-based flood fill. That's what i used to animate my "traveling wavefronts" sketch https://plus.google.com/u/0/110373479209528879659/posts/XeNRFD9Fte1 + http://creativejs.com/2011/08/traveling-wavefronts-shader-demo/

Felix Woitzel - 2012-05-29 10:52:08+0000

btw here's my latest try with float textures, http://cake23.de/cellular-fluid.html

Tim Hutton - 2014-01-03 19:40:13+0000

Together with the recent paper by +Tom Froese this makes me think: why does reaction-diffusion give such life-like behaviour?

Tom Froese - 2014-01-03 22:38:04+0000

Very nice work! In our paper we managed to get the reaction-diffusion spots to exhibit interesting behaviors, but we did not get them to evolve much by natural selection. We thought that what might be the necessary next step is to get the spots to have a relatively decoupled internal milieu, which you seem to have achieved. Now the challenge is to allow such an internal milieu to mediate the spots' interaction with their environment.

Felix Woitzel - 2014-01-04 20:01:23+0000

By the way, do you know http://www.lawsofform.org/lof.html?
I had once started with a combined effort for coupled reaction diffusion spots and a swarm of particles that actually don't interact directly but only over their trace in the environment. This one is one-way only where the particles are not yet affecting the hidden reaction diffusion patches, but the same gradient flow as in the "travling wavefronts" demo in my previous post is then added to the particle's step-forward function: http://www.cake23.de/fmx/particle-turing-fluid.html
And then, thanks to an entry on hackernews, my other more advanced experiment went viral recently and I've added a little write-up on the implementation details there too: https://news.ycombinator.com/item?id=6765029 Not only that the particles have a residual motion vector for Verlet integration now, but I've completed the feedback roundtrip and the particle projection is used in the reaction-diffusion layer too.

Why is Conway's Game Of Life Turing-complete? ;)

Tom Froese - 2014-01-05 00:34:24+0000

+Felix Woitzel Getting the spots to leave traces is indeed an interesting way to get them to spontaneously move around and to interact with each other. We have done this for models of reaction-diffusion spots. But it also works with real chemical systems, like the self-moving oil droplet. 

http://www.nature.com/news/2011/110223/full/news.2011.118.html

Felix Woitzel - 2014-01-05 13:08:03+0000

Hehe, funny little drop. Do you know any papers on designer amoebae and sensing chemical gradients for directed actions (in silico)? To enable general evolution, you certainly need some kind of genome that can mutate.
The most simple form could be this (2,3) Turing machine: http://en.wikipedia.org/wiki/Wolfram's_2-state_3-symbol_Turing_machine Maybe it is a wee bit too heavy to abstract the computational part of the intelligence but it should show you that you actually don't need a big state space like you have in natural chemical systems. The simulated selection is also just as good as its fitness metric of desired behavior factors like traveling speed, metabolism stats, or dominance over other species. Most likely it is also just implicitly implemented in the environment. I'm thinking of separated shores on a small island where individuals could flee to and recharge and one or more fighting arenas. That would be lifelike.

CC +Gerd Moe-Behrens

Tom Froese - 2014-01-06 17:18:13+0000

+Felix Woitzel For a simulation of the behavior and evolution of amoebae-like agents, you can check out the work by my colleagues:
http://www.mitpressjournals.org/doi/abs/10.1162/artl_a_00047

CC +Matthew Egbert +Xabier Barandiaran +Ezequiel Di Paolo 

Felix Woitzel - 2014-01-06 17:25:37+0000

wow, thank you!

This post was originally on Google+