Slashdot Mirror

I don't have the source code because it wasn't code I wrote. It was the pHd research project of a friend of a friend... I simply loaned server resources and provided some computer consulting. I'll do some google searching and try and dig up a reference. In brief, though, the simulation consisted of cubic 'cells' that were linked at the corners with simulated 'muscles'. The number, arrangement, and rules governing the action of the cells would be randomly 'mutate' from generation to generation. The simulation treated the cells as if they were suspended in liquid and calculated the distance the bundle of cells would move from their starting point. The organisms that moved the farthest would be more likely to 'survive' and pass their genetic makeup on to the next generation.

The simulation started with a simple one or two cell organism and was then run for many thousands of generations. At the end, it had evolved creatures that looked and swam much like fish.

This was not a unique project by the way, the guy was basing his research on an earlier, similar study. I think the earlier study was even mentioned in a documentary (or so my friend claims), so it should be easy to find with some digging. I'll google on it and see what I can find.

This paper here looks related.

Cheers,

Thad

There's also a cool video available from here.

I remember being incredibly amazed by his work when I first saw it in 1995 or so...

There already exists a sizable body of work on mixing artificial life with rigid body dynamics. Even back in 1994, people like Karl Sims were evolving their own rigid body creatures

You also want to check out Craig Reynolds (who also did lots of work on flocking)

One of the toughest challenges Natural Motion face is integrating their software in existing VFX pipelines. Many Slashdotters will already know that Linux is the OS of choice in the world of cutting edge visual effects. One issue is that Endorphin is Windows only, meaning an additional box must be placed under the operator's desk.

A more serious problem is that you cannot create your own virtual stuntmen - you have to rely on the skeletons, physiques and behaviors they supply (you cannot evolve or construct your own.) When Endorphin was first released, you needed a seperate piece of software (for example, MotionBuilder)to retarget the motion from Endorphin back onto your own characters. Thankfully though they have now implemented their own "dynamic retargetting" algorithms which smooth the workflow considerably.

To integrate it into Massive, they would need to expose the Endorphin engine as a set of libraries. However, from what I understand, Endorphin was designed from the ground up to be a stand-alone application. That after all is Natural Motion's business plan. One then wonders how easy it would be to expose an API, considering the software was never meant to be integrated into another package at all.

In its current form, Endorphin is really good for quickly generating libraries of variations which can then be fed into more traditional motion blending engines. If for example you have a crowd of people who are all falling to the ground, it can be prohibitively expensive to motion-capture enough variations to make it realistic. But with Endorphin, you just perturb the forces applied over and over again. In the end, that's their market - others are free to draw upon the underlying body of research and implement their own systems.

If it was using some sort of Genetiv algorithm with a human as evaluator, it could have been labeled as AI, but it would hardly be novel.

Karl Sims did some work on this some years ago (Im not sure if he was the first), and as you can see the result can be interesting but very abstract. Cool as wallpaper, but nothing more. BTW, Karl Sims is probably more famous for his artificial life sesearch.

If you google around you will also find work on AI generated screensavers and music.

If it was using some sort of Genetiv algorithm with a human as evaluator, it could have been labeled as AI, but it would hardly be novel.

Karl Sims did some work on this some years ago (Im not sure if he was the first), and as you can see the result can be interesting but very abstract. Cool as wallpaper, but nothing more. BTW, Karl Sims is probably more famous for his artificial life sesearch.

If you google around you will also find work on AI generated screensavers and music.

The "folks" you refer to is Professor Adrian Thompson of the University of Sussex. A paper describing his interesting experiment can be found here. It was actually a flawed FPGA chip he was programming.

Another example of this tendency of Genetic Algorithms to make use of helpful "flaws" in their environments can be found in the works of Karl Sims. A round-off error in his physics model resulted in some weird locomotion by a branch of virtual creatures.

You will find details of both examples in this entry on my Wetware blog.

Incidentally, Karl Sims wrote some papers on similar topics, including a detailed SIGGRAPH paper on "Evolving Virtual Creatures" in 3D space.

Karl Sims has also done a lot of work with evolutionary graphics and animations.
Check out his web site here especially the section on Evolved Virtual Creatures.

Karl Sims has also done a lot of work with evolutionary graphics and animations.
Check out his web site here especially the section on Evolved Virtual Creatures.

It's based on Karl Sims work, which I saw presented about 10 years ago at SIGGRAPH. His page is at http://www.genarts.com/karl/

I tried implementing something in LISP then C based on Sims' work way back but got stuck; I'm glad evolvo has emerged so I can actually play with it.

This article is interesting, however, because it moves the agents into the physical world where it isn't possible to obtain the same kind of idealized environments that are possible in silico.

This article is interesting, however, because it moves the agents into the physical world where it isn't possible to obtain the same kind of idealized environments that are possible in silico.

It had this explanation for the black oil. Life on Earth was seeded by meteorites from Mars. Well, this is an old idea, called Panspermia. You don't have to be a conspiracy nut to give it consideration.

The Mars inoculation of life on Earth came with an "Alien Virus". Excuse me? If the inoculant for all life on Earth came from from Mars, and the virus is also from Mars, how is it any more alien than any other life on Earth?

-- SilentTristero

-- SilentTristero

ps: to compiler folks, "kernel" means a small bit of highly computational code, usually with no I/O; not an OS kernel. Kernels to them are things like FFT, convolution, and so on.

for the movie files. He's been doing the same thing back in 1994, plus you can read the papers on how he did it, and they look much cooler.

This type of work is definately interesting and has produced some good results. If you are interested, definately check out the references at the bottom of the page, they are some of the defining work in this area. For your convenience I've linked up a few here (for some reason they're not linked from the actual site):

Karl Sims stuff

His Original Paper
Some cool pictures and more links

That should get you started.


Hotnutz.com - Funny

This type of work is definately interesting and has produced some good results. If you are interested, definately check out the references at the bottom of the page, they are some of the defining work in this area. For your convenience I've linked up a few here (for some reason they're not linked from the actual site):

Karl Sims stuff

His Original Paper
Some cool pictures and more links

That should get you started.


Hotnutz.com - Funny