Living Robots, Designed By Computer

12:14 minutes

Researchers have used artificial intelligence methods to design ‘living robots,’ made from two types of frog cells. The ‘xenobots,’ named for the Xenopus genus of frogs, can move, push objects, and potentially carry materials from one place to another—though the researchers acknowledge that much additional work would need to be done to make the xenobots into a practical tool.

Image of a small cluster of cells that act as a robot.
A 3D organism designed by an evolutionary algorithm and built out of living cells. Credit: Douglas Blackiston

The work was done both in computer simulations and in actual living frog tissue. First, a series of computer algorithms attempted to design a structure built out of two types of frog cells—skin cells, which are mainly static and structural, and heart cells, which can contract. A physics simulation then tested those structures to see which might have the desired function. Cell structures that showed promise were kept and randomly tweaked, while those that didn’t work were discarded. After thousands of iterations of this process, a skilled microsurgeon sculpted some of the structures out of collections of developing frog stem cells—cutting away in some areas and cauterizing together in others—to produce approximations of the computer-generated designs. The researchers then tested the behavior of the structures, each containing around 20,000 cells, in the real world, and used data from those tests to refine their algorithms still further. 

The research was published in the Proceedings of the National Academy of Sciences. Josh Bongard, a professor of computer science at the University of Vermont and co-author of the report, joins Ira to talk about designing cell-based structures and next steps for the technology.

Find out what’s happening on Science Friday…on Thursday. Subscribe to our preview newsletter.

Donate To Science Friday

Invest in quality science journalism by making a donation to Science Friday.


Segment Guests

Josh Bongard

Josh Bongard is a professor in the Department of Computer Science at the University of Vermont in Burlington, Vermont.

Segment Transcript

IRA FLATOW: This is Science Friday. I’m Ira Flatow. Later in the hour, the intersection of math and everyday life, but first, remember the 1966 movie, Fantastic Voyage? It had a really interesting plot. A submarine and its crew are shrunken down to microscopic size so they can be inserted into a person’s body into the bloodstream, travel through the arteries to repair a scientist’s brain.

But what if the future of that sort lies not in metal and silicon, but in carefully designed collections of cells? Biological robots programmed through their design to do some important function. Joining me now to talk about that, Josh Bongard. He’s professor in the Department of Computer Science, University of Vermont, and co-author of a paper this week on the proof of concept of a way to design these bio robots, published in the Proceedings of the National Academy of Sciences. Welcome to the program.

JOSH BONGARD: Thanks very much, Ira. Thanks for having me on.

IRA FLATOW: You’re welcome. You know, some of the stories about this have called these structures the first living robots. Is that accurate?

JOSH BONGARD: It’s close. We have a lot of colleagues that have been trying to build and successfully building machines out of DNA and other living components. The main advance for our work here is that we asked an AI to actually design these biobots for us.

IRA FLATOW: So they’re living things.

JOSH BONGARD: Well, I guess we could argue about that. If you zoom in to one of these biobots down to the level of a cell, it is definitely a living thing. Our little biobots we’ve nicknamed them xenobots for the moment. They’re made out of cells taken from Xenopus laevis, the African horned frog. So they definitely qualify at some level as a living system, but they are not naturally evolved organisms.

IRA FLATOW: OK, so they can’t reproduce on their own then.

JOSH BONGARD: They definitely cannot reproduce on their own. At the moment, they’re just mixtures of frog skin and frog heart muscle cells. There are no reproductive organs in there for obvious reasons. So how do they know what to do, what tasks you want them to do, and how do they do that?

JOSH BONGARD: That’s a great question. So the experiments we reported on in this study, each experiment has two phases to it. In phase one, we told the supercomputer here at the University of Vermont what we would like the eventual xenobot to do. And the supercomputer then gets to work trying out billions of different designs, where any one design is some combination of virtual frog skin cells and virtual frog heart muscle cells.

The supercomputer puts together those virtual cells in a virtual Petri dish, and then the supercomputer watches this little virtual xenobot try to move along the bottom of the dish. And it scores each one. Faster moving xenobots survive and reproduce, again, all in a virtual world, and the slow moving xenobots are deleted and replaced by the offspring of the faster moving xenobots.

We repeat this process in the supercomputer, this evolutionary process, for a few days or a few weeks, depending on what we’ve asked the supercomputer to do. In this case, we started very simple. We just asked the supercomputer to design us a xenobot that would move as quickly as possible across the bottom of a Petri dish. At the end of that day or a couple week period, the supercomputer gives us back one or a few xenobot designs.

And our biology colleagues at Tufts, Douglas Blackiston and Michael Levin, then get to work and start building these xenobot designs from actual Xenopus cells. And they then take their constructs and put them in the bottom of an actual Petri dish. And in many cases, lo and behold, the physical xenobot moved in exactly the way that the computer had predicted it would.

IRA FLATOW: Now I’m interested that you used heart cells– is that correct– in your xenobot?

JOSH BONGARD: That’s right. These are cardio myocytes or heart muscle cells. They’re like little tiny pistons. They increase in size and decrease in size and cause the overall organism to start moving, or not.

IRA FLATOW: That’s what I– yeah, I was thinking you get beating heart cells going, and that’s what your locomotion power comes from.

JOSH BONGARD: Exactly, and it turns out, though, in practice, this is not so easy to do because frog heart muscle cells, if they grow into the shape of an adult frog heart, they will sync up and they will all beat as one, which is good news for the adult frog.

But if those heart muscle cells are rearranged into completely new patterns like the ones dictated by the computer, there is no guarantee that the cells will all synchronize together. So that makes a very challenging task for the computer. We wanted to design a machine basically made out of cells that will move smoothly, but all the individual elements are, in a way, misbehaving. They’re all firing and expanding and contracting at their own rate.

IRA FLATOW: So OK, you’ve got them to move around a Petri dish. Big deal.

JOSH BONGARD: That’s correct.



IRA FLATOW: I can do that with my pencil point, right?


IRA FLATOW: And so what practical usage can you make of this, and where do you go from here?

JOSH BONGARD: Well, that’s a good question. So as you mentioned, at the moment in this first study, we just wanted to demonstrate that this technology is possible and at least how to do it for simple behaviors like movement. And it’s hard to say where a technology like this will go, what kinds of applications will come out of it.

We identified two potential applications for this technology. The first one is environmental remediation tasks. So at the moment, for example, it’s very difficult for us to identify, locate, and filter microplastics out of our waterways. There are attempts underway to build nets and boats and so on. But if we build machines out of traditional materials, like metals and plastics, unfortunately, those also degrade, and they degrade particularly quickly in saltwater, like oceans where a lot of the microplastics are.

So at the same time that some of those machines are trying to clean up the oceans they’re also degrading and contributing to the problem. So one of the appealing aspects of the xenobots is they are 100% biodegradable. So it may be in working together with some of our industrial partners, moving forward, we might be able to build very large swarms of these xenobots, drop them into our waterways, and ask them to act like little small sheepdogs. Can they find and collect microplastics into much larger masses that can then be scooped up by boats and nets and disposed of?

IRA FLATOW: I’m thinking of that Michael Crichton book where the swarms get out of control once they’re released.

JOSH BONGARD: There is definitely, as you can imagine, a lot of concern about this technology, that if we are going to build very large swarms of very small things, they could get out of control. So we want to be sure to move forward very carefully with this kind of technology. It’s important to remember, though, that there is no genetic engineering here. So if you zoom into any one xenobot, again, it’s just a simple frog cell.

Of course they might act in ways we don’t expect. But they are unlikely to get out of control, unlike some other genetically engineered organisms or weaponized viruses. So in some ways, they are safer than some of the alternatives.

IRA FLATOW: Can you mix and match frog cell parts on these to get different kinds of functionality?

JOSH BONGARD: Well, that’s actually the next step that we’re already tackling. As I mentioned at the moment, they’re only made out of skin and heart muscle cells. So there’s no reason that we couldn’t mix and match with sense organs or sensory cells as well. We might be able to incorporate or the computer might be able to figure out how to incorporate eyes and noses and chemo sensors so that the xenobots would have a better job of sensing their local environments. And eventually, we may consider including neural tissue as well, so that these xenobots can act more intelligently.

IRA FLATOW: Well, at what point does it not become its own species or whatever?

JOSH BONGARD: Well, that’s a very good question. So at the moment, these xenobots are, in essence, frogs. They just don’t look like frogs. They’re very small. They’re one millimeter across, or they’re about the size of a grain of sand. They definitely don’t look like a frog. They look like a popped kernel of popcorn. But they are definitely covered by the animal welfare regulations that exist in this country, especially as it relates to scientific experimentation. So from the regulatory body’s point of view, these are simply frogs. And there are certain things we can and can’t do with frogs in the lab.

But if we keep pushing this technology further and further, as you mentioned, the eventual machines look less and less like frogs and look more and more like something else, and we may need to work with our regulatory bodies to develop new regulations for these new kinds of organisms.

IRA FLATOW: And does each one of these have to be handmade? I mean, you’re not 3D printing them, right? They sound like they have to be put together piece by piece.

JOSH BONGARD: That’s right. So they were put together by our Tufts colleague, Douglas Blackiston, who’s a very accomplished micro surgeon. So he scrapes about 1,000 cells off a very early frog embryo, dissociates the cells so they’re all free floating in the fluid in a petri dish, and then cells will just spontaneously glue back together. And as they do, Douglas coaxes them into exactly the right configuration dictated by the computer.

So as you can imagine, this is a very labor intensive process to build a sand grain sized biobot. So we’re also looking in next steps in automating the process of manufacture as well. We’ve nailed down the automating the design side, but not yet the manufacture side.

IRA FLATOW: Can you crowdsource this? Can you put out a public cookbook of how to make your own one of these?

JOSH BONGARD: That’s exactly what we’ve done. Our PNAS paper, at least for biologists, hopefully reads like a cookbook about how we very labor intensively build these robots. And we’re hoping some of our biology colleagues will weigh in on better ways to build a xenobot.

IRA FLATOW: Did these computers come up with any designs that surprised you that you might not have thought of?

JOSH BONGARD: They absolutely did. And that’s the reason why we asked the computers to design them for us. It turns out that if I were to give you or even to give a biologist 1,000 randomly beating frog heart muscle cells and ask you to put them together in a way that we’d end up with an organism that moves smoothly is a very difficult thing to do. So the computers not only are good at this, but they come up with lots of different designs, and some of them are wackier than others.

One of our favorites was the xeno cage bot. So the computer realized unintentionally that these things need to move through a fluid. So there’s a little bit of nutrient in the bottom of the petri dish. And if you need to move quickly through a fluid one good way to do that is to be hydrodynamic. And one way to be hydrodynamic is to be hollow. So the computer came up with the cage bot, which is basically an empty cube where the struts of the cube are made from frog’s skin and heart muscle cells. And once we had that cage bot, we found out that we could also put a small pellet inside it.

IRA FLATOW: All right. I’m going to have to put you on hold.

JOSH BONGARD: I understand. I understand. Thanks, Ira.

IRA FLATOW: We’ll come back with the rest of the story some other time. Josh Bongard.

JOSH BONGARD: Sure. Thank you very much.

IRA FLATOW: University of Vermont.

Copyright © 2020 Science Friday Initiative. All rights reserved. Science Friday transcripts are produced on a tight deadline by 3Play Media. Fidelity to the original aired/published audio or video file might vary, and text might be updated or amended in the future. For the authoritative record of Science Friday’s programming, please visit the original aired/published recording. For terms of use and more information, visit our policies pages at http://www.sciencefriday.com/about/policies/

Meet the Producers and Host

About Charles Bergquist

As Science Friday’s director, Charles Bergquist channels the chaos of a live production studio into something sounding like a radio program. Favorite topics include planetary sciences, chemistry, materials, and shiny things with blinking lights.

About Ira Flatow

Ira Flatow is the host and executive producer of Science FridayHis green thumb has revived many an office plant at death’s door.

Explore More