Benito Fernandez builds robots that learn from experience

Find video of the Benito Fernandez robots at EarthSky’s YouTube page.

Benito Fernandez is developing robots that can learn and adapt. They use what he calls ‘applied intelligence.’

Benito Fernandez: Applied intelligence is a term I use to describe some of the technologies we have developed over the years, where we look at nature and see how nature adapts to the environment, evolves, learns, and then create similar devices into physical systems.

Dr. Fernandez is a researcher in robotics at the University of Texas at Austin. He talked about the utility of this applied intelligence for robots used, for example, for search and rescue.

Benito Fernandez: Robots with different skills, different sizes, different power, is a must. You want small robots, fast robots that can get into tight places, explore, go ahead, and map the environment. You may need robots that, by grouping themselves, create a structure, so that the building doesn’t collapse onto the victim or allows the firefighters to go and search and rescue the person.

Dr. Fernandez described robots that can actually build themselves in a particular way, depending on the task at hand. That’s useful when sending robots to another planet

Benito Fernandez: You may be smart enough and try to anticipate all the possible conditions and the scenarios, but it is very hard. So the best way would be to send some robots that assemble other robots with modular components. You might need robots that are small and fast or brainy, or tall, because you may need to climb, you may need to roll, you never know what you may need at the time.

Dr. Fernandez said that the robots he’s developing learn things in much the same way as people.

Benito Fernandez: One of the first things that I was trying to do was make intelligent systems. And the first thing, looking into nature, is how we adapt to the environment, how we learn from our peers, from our parents. So that led me to look into artificial neural nets, our brain, our nervous system, how it learns from experiences, what happens at the biological level, and then to mimic, in software and hardware the same things. So the device will learn and adapt, just as our brain, to behave in that certain manner. We give a positive reinforcement if they are learning the right way, or a negative reinforcement if they are learning the wrong way.

It’s hard to say whether it’s possible to build a robot that’s truly alive, added Fernandez.

Benito Fernandez: It depends on what you define as ‘alive.’ I mean, one level would be self-awareness. We’re not there yet, but it’s certainly possible. But if we compare the scale of the systems that we have today with the scale of our brains, our nervous systems, we are many years from that to happen, I would say probably decades. But if we get a large enough system, it may end up that at some point that they would be self-aware.