Print 32 comment(s) - last by drycrust3.. on Jan 29 at 2:20 PM

The lego prototypes of the evolutionary robots  (Source: University of Vermont)
University of Vermont researcher finds that robots learn to walk and behave more efficiently when its body and behavior evolves instead of being fixed

A University of Vermont researcher has created robots that are capable of evolving, much like tadpoles becoming frogs. 

Josh Bongard, creator of the evolving robots and an assistant professor in the University of Vermont's College of Engineering and Mathematical Sciences, has simulated and created robots that change body performance over time instead of having a fixed body form and behavioral traits like other robots. 

Up until this point, robots have been designed and built one specific way and are programmed directly instead of having to learn certain behaviors. But Bongard argues that this method may not produce the most efficient robots. 

Instead, Bongard has created robots capable of evolving both its body and behavior over a period of time, much like the way humans grow from babies to adults. The goal is to create four-legged, upright robots that can move to a light source without falling.  

Bongard's robots were made in many different ways. Some start out flat on the ground, or like snakes with legs while others may have splayed legs like lizards. The robots may start out in any of these forms, but the end result is that they form upright legs and know how to use them. They have 12 moving parts and are very simple-looking structures with a jointed spine, the look of a mammal's skeleton, and four sticks for legs.

The prototypes for these robots were made of Lego's, showing how the evolution of these robots would work. They were made as four-legged "robots" like in the simulation, and wore braces on its front and back legs that would tilt it. This causes the controller to look for successful movement patterns, which results in the legs going from horizontal to vertical. They would go from a reptile to a quadruped. 

"We built a relatively simple robot out of a couple of Lego Mindstorm kits to demonstrate that you actually could do it," said Bongard. 

To make the real robots, Bongard first ran 5,000 computer simulations -- each taking about 30 hours to complete -- on the University of Vermont's parallel processors to create synthetic models that move around in 3-dimensional space. Each generation of each creature then "runs" a genetic algorithm, which is a software routine that helps the creature learn different body motions such as slithering, crawling or walking. An appropriate motion is applied to each generation of the creature, giving it a proper plan to be able to obtain the goal of moving toward a light source without falling over. 

"The snake and reptilian robots are, in essence, training wheels," said Bongard. "They allow evolution to find motion patterns quicker, because those kinds of robots can't fall over. So evolution only has to solve the movement problem, but not the balance problem, initially. Then gradually over time it's able to tackle the balance problem after already solving the movement problem."

Bongard noted that robots who evolve this way learn to walk more quickly and have a more "robust gait" than those that were built with fixed bodies and behaviors. In tests, Bongard's evolving robots were able to reach the final goal of moving to the light source without falling over faster than non-evolving robots. Also, researchers found that the robots were able to attempt new kinds of challenges that were not previously given to them after reaching the light source. This may be because controllers used in the evolving robots could have maintained a certain behavior over a wider range of sensor-motor-related functions while controllers in fixed robots did not. 

"We're copying nature, we're copying evolution, we're copying neural science when we're building artificial brains into these robots," said Bongard. 

This study was published in Proceedings of the National Academy of Sciences.

Comments     Threshold

This article is over a month old, voting and posting comments is disabled

RE: Equivocating the term evolution
By MozeeToby on 1/21/2011 12:20:30 PM , Rating: 5
It's not entirely clear but I think the evolving part comes earlier in the research than the robot part. It says they ran 5000 simulations on different body plans to find optimal body plans via a genetic algorithm (I can give a rundown on what exactly a genetic algorithm is if you'd like, just let me know). Basically, they're using ideas from evolutionary biology to do the engineering work for them. You just specify the possible parts, provide a simulation and a scoring function and the genetic algorithm does the rest.

What makes the article hard to follow is that they used a second genetic algorithm to optimize the movements of each design (and I believe let the results carry over from one generation to the next). That makes it sound as though each individual robot design is 'evolving' the ability to walk efficiently, which is on a computing level true enough, but on a biologically level is closer to the way a brain develops new skills.

What they found was that robots that evolved were more stable than robots that had been designed. If I had to guess I would say that it has to do with the controlling behaviors evolving over time as the robots form changes from one simulated generation to the next. That would result in behaviors that are less specialized and more flexible than a human designed behavior would be.

RE: Equivocating the term evolution
By geddarkstorm on 1/21/2011 3:40:26 PM , Rating: 4
The best way to learn how to do something right, is to be able to sample ways that fail and gradually build up to it. That's what all learning entails. In that way, this is a seriously cool breakthrough. A truly learning machine that optimized itself via experimentation!

However... Learning is now "evolution"? Well... evolution does just mean change or unfolding, but we are diluting the word considerably; and the pictures in the article trying to tie it in to genetic based biological evolution, which this is NOT, only continues to obfuscate the meaning. This is learning. An algorithm had a hard set of parameters which it learned by experimentation to optimize the score for. Why is all learning now connotating biological evolution?

Oh well, science-terms-being-over-used-till-they-lose-their- meaning-and-breed-needless-confusion gripes aside, this is a sweet leap forward for AI and robotics research.

RE: Equivocating the term evolution
By MozeeToby on 1/21/2011 4:17:44 PM , Rating: 2
It's referred to as evolution because of the way the learning takes place. Genetic algorithms are very cool and can solve a wide range of problems, so long as you have a way to easily score a possible solution. Here's how it works:

You randomly generate a population of possible solutions. None of them are going to work, because they're all completely random, but you put them into your simulation anyway and let them have a go. If you're lucky, a few of them will hobble forward a couple inches just due to purely spastic and purposeless movements.

So, you rank all possible solutions and based on the rank, they are more or less likely to reproduce into the next generation. Depending on the technique being used, you'll generally create the next generation by combining portions of solutions from the previous (basically, 2 possible solutions produce offspring that have characteristics of both of them) and/or by through mutations (insertion, deletion, bit flips, etc). And then you do it all over again, and again, and again.

The code that describes each possible solution is analogous to DNA, the scoring function to natural selection, and the mating and mutations to... well, mating and mutations.

In this case, the researchers were using two sets of genetic algorithms in parallel with each other. One to optimize the robot's form, and a second to optimize the robots behavior. In the first, the analogy is more easily understood, each generation is a simulated robot. In the the second, the analogy isn't as apparent because nothing physically changes from one generation to the next. It's using the concepts of evolution (selection, mating, mutations) to produce a piece of software.

RE: Equivocating the term evolution
By drycrust3 on 1/23/2011 3:09:29 PM , Rating: 2
It's referred to as evolution because of the way the learning takes place.

Nonsense! As others have said, all this guy is doing is diluting the meaning of the word "Evolution". This is just an example of artificial intelligence, which should be expected with our current level of technological development.
In fact, one could easily argue it should have been expected at least 10 years ago, if not 20 years ago, so this is just another example of the theory of Evolution holding science back, not helping it move forward!

RE: Equivocating the term evolution
By Myg on 1/24/2011 6:46:30 AM , Rating: 2
The only thing "holding science back" is their utter need to disprove God and attain immortality through any means.

The LHC and murdering children for stem cells to attain these goals are perfect examples, if any are needed.

RE: Equivocating the term evolution
By kalak on 1/25/2011 1:16:43 PM , Rating: 1
You are really a troll, don't you ?
MozzeToby post a great explanation and you insist with your "nonsense!" bullshit...
The word "evolution" could agregate much more meaning than the simple "Darwin" definition. The robots software is evolving. It is learning and adapting to the surroundings. Becoming better robots (robots with a larger chance of survival). Is it not evolution ?

By drycrust3 on 1/28/2011 6:03:21 PM , Rating: 2
You are really a troll

No, I don't consider myself a troll, I consider myself as a champion for good science. Why should proponents of the theory of Evolution get an exemption from the normal protocols of science? No other field of science is given an exemption, and you certainly wouldn't allow Creationists to postulate theories without some credible proof as to their validity, so why should Evolutionists be given an exemption? If, as they claim, their theory is so scientifically sound and so well proven that it should be regarded as a fact, then they should be the first ones following the rules, not the first ones breaking them.
The word "evolution" could agregate much more meaning than the simple "Darwin" definition.

If MozzeToby wants to postulate a different theory from what Darwin postulated, then he is entirely free to do so. The problem is that when we talk about "the theory of Evolution" then we have to talk about what Darwin postulated because that is the only theory of Evolution that counts. If we add bits to his theory then it isn't his theory anymore.
This article highlights the fact that what people perceive as the theory of Evolution and what it is are two different things, which surely is a sign that what Darwin postulated is seriously out of date with what science has revealed to us about life.

The robots software is evolving.

Is it? Gathering data from sensors about your body and your environment, storing it, processing it, and then deciding on how to behave isn't software evolving, it is just an example of a computer program doing what it was intended to do!
As I said above, if you seriously believe the robot's software evolved then prove that it has evolved.

By mmcdonalataocdotgov on 1/24/2011 7:14:46 AM , Rating: 2
So basically the hardware engineers were using standard product R&D techniques, and the software engineers designed the control systems to learn the best configuration. None of this is new in the business world - it is mostly quantitative business analysis and change management. I guess if you have only ever been in academia, it is fun to think of it as evolution.

"When an individual makes a copy of a song for himself, I suppose we can say he stole a song." -- Sony BMG attorney Jennifer Pariser

Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki