Co-Evolving Robots At Brandeis

neck jones pointed out this site titled "Towards Fully Automated Design of Real Robots" at the Brandeis Dynamic & Evolutionary Machine Organization Lab which dropped my jaw. As neck says: "Whoah." Anyone who can summarize their work by beginning "Start with a set of simple bodies and set of random brains" and go on to describe automated, automatic fused depositon manufacturing already has my attention.

Comments

Automated whose-it what's?

[tcd004]

This will be really cool till Linda Hamilton crushes it in a hydraulic press.

tcd004

Here's my Microsoft parody, where's yours?

Really Cool!

[Utoxin]

This kind of stuff is what really gets me excited. Imagine the potential of this stuff if we sent a small set of evolving robots to a distant planet. They could have a database containing information on how to build various types of robots, and also, for eventual use, information on how to create humans, using a digital copy of human DNA, or frozen fertilized eggs.

Imagine it... Robots being used to help us colonize other worlds. Is that not /cool/?
--
Matthew Walker
My DNA is Y2K compliant

Re:Really Cool!

[gentle_giant]

Yeah, sending robots to a distant far-away planet with frozen fertilized eggs would be cool, but what if it worked? Would it really ensure the survival of our civilization? Or would these new humans be a totally different race because they had no cultural ties whatsoever to us? Would they just die from depression because they had no older human to care for them, teach them, etc.? And what is more important to us: the survival of our species or the survival of our civilation (art, literature, way of life, etc.). There is so much we don't know about ourselves that it would be silly and perhaps even cruel to send unborn members of our civilization on such a mission. I mean, if you were one of these babies, wouldn't you ask yourself, "Why did they send me? Didn't they want me back on Earth with them?" Just a few thoughts on the topic.

Makes you think...

[c=sixty4]

Wasn't this exactly the sort of thing Bill Joy was warning against? "Beware the dawn of the Metal Age!"

Seriously, I'm a bit skeptical to the idea of construction by essentially throwing dice.

Re:Makes you think...

[chialea]

well, I worked next to their lab a few years ago (and one of my friends worked in the lab, but hung out in mine -- more room) and they were working on this sort of thing then. and I have to say, I've never seen anything even vaguely pretty come out of this, but what comes out does fit the specs. the lego cranes can really lift the load, the little running guys stop running into things, whatever.

when you start going automated, you get stuff which is ugly, if you don't specifically program esthetics in there. you get things that are inefficent, if you don't specifically program efficency in. you get things which are just amazingly silly, if you don't do your limiting conditions (or your math!) quite right. this will give you exactly what you ask for. right now it's "just" a matter of learning how to specify things, and more effecient ways to "breed" them and prune out the tree. an interesting technology to be sure...

Lea

Cool step in Robot evolution!

[MAXOMENOS]

This work out of Brandeis is an implementation of the thought-experiment done by Valentino Braitenberg in chapter 5 his book Vehicles, which outlines experiments in evolution of simple robots. The main differences between the Brandeis work and Braitenberg's experiment are that the robots are being constructed to particular practical ends and most of them are simulated before they are built. Damn. Wow. Well done!

The Second Amendment Sisters

Domo Arigato

[Anonymous Coward]

Domo Arigato, Mr. Roboto
Mata ah-oo Hima de
Domo Arigato, Mr. Roboto
Himitsu wo Shiri Tai

You're wondering who I am - Machine or mannequin
With parts made in Japan, I am the modren man.

I've got a secret, I've been hiding
Under my skin.
My heart is humn, my blood is boiling
My brain IBM.
So if you see me, acting strangely
Don't be surprised.
I'm just a man who, needed someone
And somewhere to hide
To keep me alive - Just keep me alive.
Somewhere to hide to keep me alive.

I'm not a robot, without emotion
I'm not what you see.
I've come to help you, with your problems
So we can be free.
I'm not a hero, I'm not a saviour
Forget what you know.
I'm just a man whose, circustances
Went beyond his control - beyond my control.
Beyond my control, we all need control.

I am the modren man, who hides behind a mask
So no one else can see, my true identity

Domo Arigato, Mr. Roboto, Domo (Domo), Domo (Domo)
Domo Arigato, Mr. Roboto, Domo (Domo), Domo (Domo)

Domo Arigato, Mr. Roboto [Repeat several X's]

Thank you very much, Mr. Roboto for doing the jobs that nobody wants to
And thank you very much, Mr. Roboto for helping me escape just when I needed to

Thank you - Thank you. Thank you.
I want to thank you, please thank you.

Oh! Oh-ah-oh!

The problem's plain to see
To much technology
Machine's to save our lives
Machines dehumanize

The time has come at last.
To throw away this mask.
Now everyone can see.
My true identity.

I'm Kilroy!
Kilroy!
Kilroy!
Kilroy.

Other related stuff of interest.

[Matt2000]

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

Hmm..

[technos]

1. Introduce the idea of survival through predation.
2. Add some virtual physics.
3. Slap the simulation in the world's biggest cluster.

4. Sit back and see how long it takes the simulation to nuke itself!!

Now seriously.. All this amounts to is simple expansion of neural net evolution technique... We're still limited by the brains we can give the little virtual monsters..

Re:Hmm..

[technos]

They are certainly allowed to evolve! Unfortunatly, there are a set of well-known linits (I suppose) that govern how far a net can evolve.

It's not the machines, it's the huumans that designed them.. We're simply not smart enough to create a neural net capable of evolving past a certain point... The net stagnates on a set of variations and refuses to move..

I'm not an expert by any means, so any further explaination should be left to others.. . .

Hrrm...

[Signal 11]

Sounds like they've managed to duplicate the intelligence of an average AOLer. What an advance in artificial stupidity - I'm impressed!

Try it yourself

[chuck]

The whole simulation of robots angle reminded me of this site: http://sodaplay.com/.

It's a java applet where you can design some silly little robots in 2-D, and see how you can make 'em work. No neural networks, or real-world synthesis, but hey, it's cool!

Don't moderate Signal 11 down!

[Anonymous Coward]

Signal 11 is really a 98 year old deaf and blind black woman who lives in a geriatric home in Kansas. Slashdot is everything she's got, so please go easy on her.

Rudy Rucker

[Fool@Work]

This looks very similar to ideas found in Rudy Rucker's fiction. Those interested in some good, entertaining science fiction type stuff might want to check out his books, especially Software and Wetware. I'd almost call his books "cyberpunk", but they aren't dark and gloomy.

Fool@Work

Very cool

[sbeitzel]

I want way more detail on the "printing" process. Man. That looks really cool. I think it'd be awesome to be able to take an arbitrary 3-D model and "print" it as a fully-detailed plastic model.

Re:Very cool

[GossG]

I didn't read the article, but what you describe is available. You start with a tank of clear liquid polymer waiting for (visual? UV?) light to cure it. You shine lasers on it from various directions, and where the lasers intersect, there is enough energy to cure the plastic at that point.

I believe I saw a TLC show where they were "printing" a 3-D lifesize HOLLOW model of a guy's skull based on a CAT scan, so that the surgeon could plan his operation around landmarks on the INSIDE of the skull.

Shorter Summary Available

[grnarrow]

If anyone wants to read a shorter, less technical summary of the work, the current issue of New Scientist has this brief article available.

It certainly sounds like an interesting frontier, but I wonder if it will ever really be practical.

grnarrow
www.bottomquark.com

Re:Rapid Prototyping (ex Re:Very cool)

[Chep]

Or even better :
http://www.machpro.fr/machines/proto/default.htm

(former link was on CAD systems with the required "drivers", not the machines themselves)

sorry there aren't much pictures there, but at least you have entry points to look further on the web.

U of D Spring Lecture Series

[Life Blood]

I'm a Univ of Delaware Mech Engineering grad student and we had a talk on this from a related researcher earlier this year. It has some cool potential in a lot of areas, but also some strong disadvantages.

Basically a modular robot is cool in that it can adapt to situtations, have redundacy in case a module fails, etc. Makes for a great exploration units.

The main problem with them is that they're a bitch to control. The processing demands rise with the square of the number of modules, so they get sluggish pretty darn fast. They also are more inefficient than a committed robot and can have problems with local weaknesses. Basically a bad configuration can easily overload one module and cause failure of the whole robot. Preventing that takes even more processor time to test possible configurations, creating a wicked cycle.

Re:U of D Spring Lecture Series

[chialea]

ok, here goes the second try, since this didn't seem to post before, for whatever reason (even though I got the confirmation screen!). hmm. bug?

in any case, I was asking whether that was Dr. Agrawal who gave that talk. my guess is that it was, since he worked at PARC last summer on PolyBot, and that seems to be what you're referring to (say hi to him from me, if you see him, if you would)

this is different than modular robotics is a few ways. this is "evolving" robots for one specific task. controlling them (in general) won't take nearly as much processing power, since it's controlling itself in a functioning (and efficient, if you select for that) way with an evolved brain. modular reconfigurable robots take so much processing power if you control them in an extremely general fashion, because they're so flexible. one robot, infinite tasks and infinite ways to accomplish it. TMTOWTDI^n, really.

I have to agree -- they're really a BIG pain to control, because of force constraints, shattering acrylic, and such (a lot of this is fixed by new hardware). it basically involves a very large amount of experimentation and tweaking, at least the way I end up doing it. :) it's a very interesting controls problem to do this in general, and perhaps the DEMO lab's techniques could be applied here. I don't believe there was any work in this area done on PolyBot, but it could have interesting results.

Lea

Okay, this one deservs a big WOW!

[Effendi13]

Besides the program's innate ability to recreate star constallations, I do have to say this is interesting. The whole concept may even lend proof to evolution in a very simplistic way. If we gave a brain to a peice of silly putty and let it run around, would it eventually be a lizard or a snake? If we put it in water, would it become a fish? It is even becoming appearent that thoughts and decisions could effect evolution, wouldn't you say? If we program the peice of silly putty in water that the faster it swam, the better, and it randomly changed parts of it's body, when it becomes faster, it locks a change until another change makes it faster... well, you could actually write a program to simulate evolution. Hmm.. just a muse I guess, but doesn't it make animals look simple?
-Effendi

For Those who fear "The Matrix AI"

[Dolio]

As we all know in the movie The Matrix, AI somehow came to destroy humanity. This is what many of us fear AI will become. I'de like to point out that we usuaily get out of technology what we put into it. If you are Micro$oft, you put crap into your os, you get crap out of your os. If we design AI as a weapon, I have NO DOUBT that AI will destroy us. So, I believe we should take great care to avoid creating destructive systems, but rather systems that are friendly. Construction systems like the NASAs robot snakes, automotive safety systems, explorers, harvesters, educational systems, and such.

AI devices need not be concerned with power or money of world dominance.

And I don't believe Microsoft is Evil,
if only they cared about their customers,
like that care about their stockholders.

$0.02 - Please overlook my spelling and gramatical errors... you get the idea i think.

Peace
L8r
Dolio

Re:For Those who fear "The Matrix AI"

[The Evil Beaver]

If Microsoft made robots with AI for the US Military, to replace each and every soldier, the United States would be taken over by Cuba. Think about it, these robo-soldiers would at random reboot.
The story of the future in both The Matrix and Terminator (both of them) has robots who have taken over the world. Starsiege and Terminator have master brain robots that made sure they wouldn't be unplugged.
It's only natural that one day, the world governments, if not stopped, will start using robot soldiers with AI. And if they build a superintellegent master machine with better AI than anything else, we are doomed. I agree with robot soldiers, BUT it must be every bot for itself.
BTW, it would be interesting to design machines that evolve. Four components - Body, brain, mine (to get raw materials), and factory (to convert materials into energy and add/repair to bot).


When the pack animals stampede, it's time to soak the ground with blood to save the world. We fight, we die, we break our cursed bonds.

Re:My own personal experience with co-evolution

[technos]

I'm laughing my ass off, but apparently no one else took the time to peek inside ILUVYOU.txt.vbs..

Karl Simms did this in 1991

[chrismckinstry]

One of the very first things I downloaded from the web in 1994 was this mpeg file of Karl Simms' evolved swimming fish. It was a hell of a download at 10mb over 14.4Kbps, but it was worth it! I am certain this is a hardware version of Simms software project (he did it on one of those giant Thinking Machines Hypercubes that the Defense Department loves so much)

The video shocked me and changed my life. It's a wonderful video to show people that don't believe in evolution. After seeing this, you really have no IGNORE hard to remain IGNORant of the power of evolution.

I can't seem to find a copy of the paper behind this video, but it is equally shocking. Try to find it yourself, it is definately worth reading.

"Artificial Evolution for Computer Graphics", Karl Simms, Computer Graphics, Volume 25:319-328, Number 4, August 1991, SIGGRAPH'91.

I remember at the time thinking it would be great if the simulated hardware in Simms' simulation (how many Simms would a Simms Sim sim if a Simms Sim could sim?) could actually be constructed... you could evolve a robot in a machine, and only actually build it after it was perfected.

Looks like that is exactly what going to happen

-- Win a MINT copy of BYTE MAgazine Issue #1 = September 1975

Sounds Familiar?

[Ralph Bearpark]

"Start with a set of simple bodies and set of random brains"

That's a description of Slashdot, isn't it?

Regards, Ralph.