Introducing: Evolutionary mobile robots

The image on this post isn’t from Dr. Fernandez’ lab. It’s from Wikimedia Commons … heralding new robots to come?

Benito Fernandez is an associate professor of Mechanical Engineering at the University of Texas in Austin. Originally from Venezuela, Dr. Fernandez is an expert in Applied Intelligence, which involves the use of different technologies to create intelligent devices. I spoke with him in early August about what he calls “evolutionary mobile robots.” Here are a couple of excerpts from our interview. More with Dr. Fernandez coming soon.

Jorge Salazar: What is an evolutionary mobile robot?

Benito Fernandez: Right now what you will find heterogeneous robots in our lab. They’re not the same. They might be of different sizes, of different sensors, that handle different things, different skills. So if you have a group of robots, how do they learn from each other, share information, learn about the environment, or coordinate action? The evolution part is two-fold. The robots can evolve mentally, so after they experience the world, they reconfigure the way they view the world, or physically, the robots can reattach themselves, or reconfigure themselves physically, so a robot in the next reincarnation or generation may say, I want to be faster or I want to be stronger. Given a particular problem or application, there might be an optimal solution of robot structure that would be more suitable for the problem at hand.

JS: Can you tell me more about what kinds of robots you have in your lab?

BF: We have several robots of different sizes, they move around in the environment, they map the environment, and they talk with each other. We have three robots on bomb detection and disarming, but we also have several robots that can do mapping and some of the visual world. As the information comes from the robot, a map is being generated in real time of the world. So you’re not there, the robots are there. From the maps they make, the human can see what the environment looks like, and based on that information, plan a rescue or something like that.

JS: How did you develop these robots?

BF: What we do is look at nature and see how nature does its thing and then try to design a circuit or software implementation of that. We know humans learn through neural networks. So I created an artificial neural network. Now the robot can also learn from the experiences they have.

After the neural net, the next thing is, how do I express knowledge so that a human can understand? You talk about things like, if it’s hot, but not too hot, turn the air conditioning on. So what is hot, and what is too hot? This is not a precise, is the temperature more than 82.3 degrees. But that’s why we convey knowledge. I’m using a language that’s not very precise, mathematically. So that took me down to fuzzy logic – dealing with this impreciseness of language. Then I tried to put the two together, fuzzy logic as a neural net and vice versa.

JS: Where does evolution come in?

BF: I started realizing some of the limitations of these tools, and it eventually led me to evolution. The human brain forms interconnections within the first five years. And after that, the plasticity of the brain is severely reduced. So the potentiality of what a brain can do is is pretty much set by five or six years.

So if that potentiality isn’t good enough to solve the problem, then you have to basically make a new brain, that evolves. So the systems that we build are neural nets that also evolve. They evolve from one generation to the next, they grow as the problem requires and eventually come out with a solution. If we look at history, how animals and plants have evolved due to the environmental conditions at the time, the same things happen with these robot systems.

JS: But how exactly do robots evolve?

BF: In the last eight years, I’ve been also working with what are called artificial immune systems. One of the things about neural nets in general is that you need a teacher, somebody that will tell you, this is how you do it, or this is good or this is bad. But if you send a bunch of robots, say to Mars, you may not have a teacher there at all. So the robots have to figure out things for themselves. The only thing that I could think of in nature that does the same is the immune system, where over millions of years, it’s still around. If they find a virus, they figure a way of fixing it, by creating anti viruses. So I took a look at how the immune system works and tried to build similar things, combined with neural fuzziness. Basically, over the years, I created a bunch of tools that I put under the name applied intelligence, which puts all of these things together and try to solve real problems.

Jorge Salazar