Abstract
As it started to consolidate, the world of computer science was thinking in that way quite a lot, but pure engineering was not thinking that way very much. Because we're so enamored of fast electronics and being able to build incredibly sophisticated control systems in electronics, there didn't really seem to be a need to go in that direction. You made the mechanical system as simple as you could for it to do the job so that it was easy to control.
Once you move into the world of deformable, nonlinear, anisotropic materials, that type of traditional approach doesn't work very well. Now you suddenly are up against computational demands that are just ridiculous. That's where I think it started to get some ground. Soft robotics was able to capture that whole idea of morphological computation. I still avidly believe that's the way it works. We call it neuromechanics when we're doing neuroscience, but who can point to the killer app, if you will, the robot that demonstrates this to the max, where you could not possibly have had it do that job without having morphological computation?
I think of myself as a scientist with some engineering tendencies. I'm not formally trained as an engineer, but I love engineering and I think that it's probably something that I've done since I was a very young child. I think very much like an engineer, but my profession has always been science, and they're not always the same thing. We can certainly talk about some of those differences if you're interested.
Research engineers, even though they are more like a scientist and they're going into areas where they don't have the answers, they're still being driven by finding an answer. It doesn't necessarily have to lead to a new truth, it just needs to be a good, better, more efficient, novel way of doing something. Scientists tend to work in a world of questions and unknowns in a way that I think is a little uncomfortable for some engineers.
I'd always had in the background of my thinking that it helps to be able to build something to be able to understand it better. Once you try to build it, you realize all of the things that you'd failed to think about when you were studying the thing.
We were studying the caterpillar for variety of reasons. It's a little obscure, but it turns out that it has huge technical advantages. We can isolate the nervous system and keep it alive, we can record from neurons very easily, and we can identify them. Its entire anatomy has been well mapped out, we know all the muscles; there are 2,000 muscles in a caterpillar. That's 10 times what we have in our body, They're fascinating creatures. We were primarily interested in using those technical advantages to study how the nervous system functions.
Now, it's only really worth studying a nervous system if you're interested in the peculiar things that nervous systems do, rather than using the nervous system to understand how liver cells function. You study the brain because it does unique things that brains do. The more we were studying the details of the biochemistry and the synaptic connections in the nervous system, the more we had to stand back and say, “Okay, what does this mean for behavior?” Behavior is tied to the way nervous system's function. And the more we looked at motor control and sensory input, we were saying, “I don't really understand how this thing gets around in the world. How does this caterpillar actually feel? What does it sense? What does it need to sense? How does it control its body?”
Once you come to that question and you recognize that, well, we know how to control some sorts of machines and robots, and you realize that it doesn't apply to a caterpillar, it doesn't really apply to a worm. We started to realize that the experimental animal we were working on was actually a bit of a conundrum, that we really had no idea how something that had unlimited degrees of freedom essentially—it can turn itself into a knot, it can deform endlessly—how on earth does it control its body?
A traditional answer from an engineering viewpoint might have been, “Well, it dedicates more of its control systems to controlling its body. It has too many degrees of freedom, so it has to have lots of muscles, and lots of control, and a complex amount of a system.” We thought about that for a while, and we looked at it, and we said, “Well, wait a minute, it turns into a moth, and a moth has an articulated skeleton with joints.” If we can compare basically the nervous system in those two stages of the animal's life and ask the question, are the nervous systems very different?
Of course, apart from the highly sophisticated sensory systems, that moths have the basic motor control, they don't have more neurons, they don't have a huge number of neurons compared with a caterpillar, or vice versa. That made us realize that the caterpillar might be an interesting way to look at controlling deformable machines, because somehow they can do it without having bigger brains. In fact, if you look across the animal kingdom, with the exception of cephalopods, our friend the octopus, invertebrates have smaller nervous systems. I'm not going to say they're simpler, but perhaps they are, but they certainly have fewer neurons.
Somehow animals have solved the problem of massive degrees of freedom without increasing the size of the nervous system and the number of connections. That made us very interested, and that's how we started to get into soft robotics. At the time, it was not really called soft robotics. It was biomimetic robotics, bioinspired. Many groups were doing soft materials in robots, but it hadn't consolidated into a field. We realized that by studying the caterpillar we might be able to provide some insights for how to engineer some deformable machines.
We started doing that in the early 2000s. Many people have built machines, robots, to try and understand how animals work. Way back in the 1950s, Grey Walter in England was doing this with his so-called turtles, and Ken Rhoda, who was a professor here at Tufts University from the 1930s until 1970s, he built a robotic cockroach, lateral roboticus, I think it was called. We came from thinking about the animal, recognizing that this was not something we knew how to do in the engineering world, and trying to put the two together.
I think the issue still is that the jury's still out until we can show there is morphological computational control of a machine that could not have been done without that approach. We're not quite there yet. There are lovely, sophisticated models, reservoir computing, and all sorts of things where you could replicate central pattern generator approaches without having a central pattern generator. A lot of that's been done, but what I want to see is a robot that can walk around in a complex world using these concepts and principles, and I think we're still a little way out yet.
We were saying, “Look, there's all these folks who are doing this work and they don't really have a forum in which they can present it. They're scattered across many, many, many journals, many, many conferences, they're in biology, they're in computer science, et cetera.” We sat down and we said, “What do we think the field will need to move forward?” It was a way of gathering our arms around all the different ideas, even though they were not necessarily roboticists.
We started out, with five or six fields that we wanted to cover, computer science being an important one, partly because the idea of morphological computation was really strong in that field as well. Material science, absolutely. Soft robotics is expanding the toolbox of materials available to engineers from basically rigid things that we can use, rigid mechanics, to just about every material that you can imagine including biological ones, proteins and fats, and things. We needed material science, computer science, we needed traditional roboticists, absolutely, and we needed control theory.
The other big area that I wanted in there was biology, and that's obviously because of my bias, but I wanted to make sure that by starting the journal, and maintaining a position with the journal, and keeping it going, we could help to incorporate and cross over some of the biology into engineering and vice versa.
I think that's been reasonably successful, but not as much as I would like because I think it's difficult to have most biologists think of themselves as studying ideas that are relevant to robotics. Soft robotics is one of the few areas where I think they do, but most biologists do not. I know that you've been talking to some of my colleagues here who work in biology and morphological code writing, and so there are a few people like that, but they are thinking incredibly deeply and broadly, and I don't think every biologist thinks that way.
The journal has added more types of expertise as well, because we needed to have more expertise in sensing, for example, electronics, structural electronics, and flexible electronics. We have gradually added more associate editors in those areas and tried to put our arms around that. The field of soft robotics is not unique, and I think it exemplifies embracing ideas from all sorts of disciplines, probably every discipline that I can think of actually. It's important that we don't dismiss ideas and areas of science out of hand, because we have to be flexible.
I don't think it would've been possible or it would've been probably ill-advised as a young assistant professor to have gone off and done that, because the way the whole evaluation system works at tenure, and the pattern in maintaining your position in academia, you need to be accomplished, and people have to recognize that, because that's the way it works. Just being a clever ideas person and pursuing things that are going to be risky is difficult in the career structure that we have.
It's incumbent on established people, certainly people who feel secure in their positions, to take big risks, and to take big risks on supporting junior people in the field. I have to say I have benefited enormously from all the people who have come to my laboratory because most of what has been really fun and exciting that's been in my laboratory has been from the people who joined my laboratory, not from me. I can throw out ideas all the time, but they're not going to land all the time.
But the people in my laboratory are those who push things forward and say, “Hey, I really want to build a robot out of foam.” Why? Let's talk about it. Let's go there. I think what we need is for our senior personnel to be really open to that, to support younger people being able to pursue these new ideas, and then pioneer them together. I think that's probably the solution we have because it's risky.
Another thought occurs to me in terms of the history of the field. A lot of the current field of soft robotics was formed 10 or 12 years ago from a huge influx of funds both in Europe and in the United States. In the United States, it was mainly through programs funded by the defense department. In Europe, I think, the European Commission was funding a lot of this. They put a lot of funds into trying to develop new types of soft robotic technology, and there was this massive innovation. This is where we start to get jamming systems for doing STIFF-FLOP type probiotics.
We get incredibly clever types of pneumatic systems, the new nets, and things that can move incredibly complicated ways just because of the way the pressure is distributed, had people building actuators that were just little explosion machines to make the robots jump and move around. We had the different types of actuators, shape memory alloys, and twisted strings, and electroactive polymers.
We had this massive innovation because the whole plane of ideas was open, it hadn't been explored before, so it was easy to fill up with novel ideas. What has happened since then is that there have been far fewer jumps forward, and that's inevitable. It's easy to come up with all the new ideas when nobody's ever done this stuff before, everything's novel, everything's fun, everything's interesting, and you can explore it.
I think we're at the phase now in the field of soft robotics where we're starting to figure out what works, what doesn't, what's a good plan, what's the details? It starts to come down to really good engineering so you can make something blow up and twist. How are we going to use that? What are the limitations? How do we make it better? How do we apply it to a particular circumstance? Now we're down to the nitty-gritty details, which can be a little frustrating to some of us in the field who want to see the next big paradigm shift. We want to see where's the next really cool idea. Whereas a lot of what's being done now is the hard work of making good engineering.
I think we've moved to a slightly different phase and it's going to take really creative people to make those new paradigm shifts and come up with something completely different. I hope that that happens. It's going to be less common than it was, in my opinion, although there's more people in the field.
We're in a difficult stage, and I think what the field really does need at this point is, as we were talking about earlier, not just a demonstration of how morphological computation works, but actually a demonstration of machines that can take all of these cool ideas and do something we couldn't do before. Most robots or robot technologies that get published in society conferences and even in journals don't go very far, they end up on a shelf.
It really is upsetting and disappointing because sometimes it then needs another engineer to reinvent it and say, “Oh wait a minute, somebody did this 10 years ago, and that's really cool, but nobody's picked it up in the meantime.” I think the field would benefit enormously from there being a well-accepted clear application that we are able to fulfill and that will make society better, and we haven't quite gotten there yet. We're still inspiring the entertainment industry, which is one of the things we've managed to do quite well, but we've not yet made that machine that couldn't be made any other way.
In answer to your question, I think we do need all those ideas, they're wonderful. We probably at this point, as a field, would be well advised to focus on making the machines that will save people in an emergency situation, for example, where it would then become much more accepted that these novel technologies and ideas have a value and are important. Ultimately, we're mostly funded by people's tax money, occasionally by companies, but usually by tax money, and people need to see that this stuff is worthwhile. There are many problems in the world, and if someone sees that there are folks starving in the world or being buried in an earthquake and not being saved, and why the heck are we funding Trimmer to do work on caterpillars? We need to fill that gap somewhere.
We get a tremendous number of articles submitted to Soft Robotics from China. We also have many articles from Japan and from Korea. They're very important. We're also getting articles from India, we get quite many from Australia, and some from New Zealand, and a few also from South America. This is now broadly representative of the way those economies have invested in their research and development.
The other part of this that I think is great is not just the investment of what governments and agencies are doing around the world. It's been the people who have come out of this. One of my greatest pleasures is not just the research articles that we publish, but also the fact that we train people who go on to do great things in a field because that's where you have an exponential influence. The young scientists who you train are what sustain the field. It's been a great pleasure to me to see people who were originally either graduate students or postdocs, who were part of that early development of the field, who are now establishing their own laboratories and are training their own PhDs, master's students, and also undergraduates. They have spread through the world.
Many folks came to the United States into Europe, and then moved back to their homelands and established their own laboratories away from Europe and America. It used to be that people would say, “What's soft robotics?” and that's not true anymore, it's just accepted, it is a field. I don't have any great insights into the way in which governments have operated to do this, I think that they tend to be very driven by what they think is going to be an advantage or what is going to be useful, but it's been an interesting journey, it really has.
Space exploration is another one. You don't want people in a space capsule with an unpredictable piece of technology. The question really comes down to, where are things going to go, where are the barriers to the technology being accepted the lowest and the easiest to adopt? It's a risk-reward thing.
There are some industries where the barriers are quite low, and the rewards are really high that are ones we don't normally think about very much. I don't want to get into it too much, but there is a lot of interest in personal robots for the sex industry. Now, that's not something we as engineers want to be thinking about very much necessarily, particularly if we're inventing this stuff, but that has already reached a point where there's so much money in the field, the risk in terms of things going wrong is not that high, and so the barriers are quite low. It has taken off in a big way.
The areas that I think we are more likely to see something that really benefits humankind in a much more interesting way would be in the area of search robots. Putting things in space would be great, medical would be really good, but I think we could, very quickly, if we put our mind to it, make soft robots that can find people in dire situations; for example, hurricane damage, buildings that have collapsed, war situations, where we need to find people quickly in extraordinarily complex situations. Right now, our traditional robots have been terrible at that.
I don't think there's an example of a traditional robot saving somebody in that situation. Those robots are also extraordinarily expensive, so they're not available to most people in the world and there are very few of them. Imagine a disaster situation in an island community, where people are buried after an earthquake. It doesn't help to have a $6 million humanoid robot that can open barrels and drive cars. What we need is something else, and I think it would be relatively straightforward.
This has been a dream of mine, because working on caterpillars and being able to make something that crawls, it seems to me we can make survey robots incredibly cheaply that can burrow through debris fields and underneath buildings. Small ones would probably cost a few hundred dollars each, and we could have an emergency trailer full of tens of thousands of these things. You take it to the emergency site, let the robots burrow down, and they have sensors on them, and you can record where there are life signs, and you can now have an opportunity to go and get that person out.
It's a very specific example, I know, but I think that we are missing an opportunity if we don't try and invest in developing that technology. It's within reach. This doesn't need us to have completely new technology that's not yet invented. I think we can do it with the technology we currently have. We just need to engineer the heck out of what we have. We could do that within 5 years if there was enough investment in that.
We've had people working on this for quite a long time; it's used a little bit in some manipulator systems, a little less so in locomotory robots, but I think we're at a point now where there's far more acceptance of what I would call learning approaches to control. Artificial intelligence (AI), effectively, where we don't necessarily have to define all of the parameters for our control system, know it ahead of time, and build the system and make it work, and then be adaptable, that's the other problem. I think there's more acceptance in the engineering field of learning approaches.
I remember very early on, when we were trying to develop some robots that I got very excited by the idea of, “Oh, we have some sort of nerve net system, a generic algorithm that will run, and we'll figure out the best solution, and it will make this thing work.” We even tried to get a grant to have many of our little software and robots available for people to drive around on the internet so that we could collect information that would allow us to understand how people controlled the robots so that we could control the robots better. Many people said, “If you have to have it learn how to move, it means you don't understand it well, and that you really can't design and build a machine that you don't understand.” It was very frustrating because it was a resistance, the idea that we don't have a predefined set of equations that we can apply to this.
I think we're past that now. I think that the areas that will help is for us to develop smart systems through AI that we don't necessarily completely understand how they work, but that will quickly train our designed robot to work properly. All we need to do is say, “We want task level control. We want this robot to be able to wiggle through here and do this under these circumstances, change this locomotion mode. If it encounters a wall, it needs to be able to get out of that situation. It needs to be able to do this under these circumstances. They're my criteria. I don't care how it does it, give me a control system.”
I think that that's going to work. Eventually, of course, we would love to know a little bit better how to design it more, but we can then have an AI that helps us to design that control system. I think that that's going to be a tremendously exciting area, and hopefully that is going to garner some serious attention and investment.
Do we want to mimic muscle? The so-called artificial muscle, should it have the properties of muscle? Possibly not. Muscle is limited. I work on muscle, and I know that many people think, “Oh, muscle's the best actuator for a soft thing. It's perfect.” But it really isn't. It's unpredictable and has many limitations. I think our so-called artificial muscles should pick the things that they want to be able to do well and those are what we should focus on. We shouldn't be trying to copy and mimic the biology.
Now, there is a flip side to that, which is that sometimes the biology gives us answers that we would never ever have explored and thought about ahead of time. I think we can take inspiration from the concepts that are embedded within biological solutions. A muscle, for example, is composed of many cellular fibers that interact at the molecular level that are busy converting adenosine triphosphate into movement, and they do it through an incredibly complex mechanism. The reason it works so well is that it's extraordinarily well aligned and organized.
It's a polycrystalline array of molecular motors. Maybe that's one of the ideas that we can grab and say, “Look, every one of our little artificial muscles, whether it's an electroactive polymer, or a shape memory alloy, or heat activated polymer,” or whatever else, instead of thinking of that as a bolt material that should be an actuator, we start to say, “No, let's put together lots and lots and lots of these in a way that allows us to achieve what muscle does.”
I think we have to balance bioinspired and bioinformed carefully, and make sure we're using the biology in an appropriate way. All my colleagues in biology will tell you that everything in biology is a massive compromise. Nothing is optimized for one thing, or even two things. We definitely should not be just copying. One of the things I'm interested in doing is skipping over making artificial muscles, and simply saying, “Why don't we just grow muscles?” We already have them, and they grow. The idea behind this is that we could potentially become soft roboticists who are also bioengineers, and that's also now a part of soft robotics.
If we know enough about the way tissues are determined and how they get formed from the original cells, we should be able to engineer that to make stuff we want and not make stuff we don't want. We're quite a long way from that. Many people have attempted to do it. Most of the time, what they're doing is taking cells that should be muscles and putting them in a dish and hoping that we can figure out what's not right with them and fix it. But it's a long process, and I think probably a more directed approach would be for us to understand how a tissue gets made from its original cells, and then be able to engineer that because now we fully understand it. It would take massive investment in looking at biology to make engineered biological machines.
It's a different approach. We're not just taking the biology and using it, we're understanding the biology to be able to develop new things. Mike Levin, who works in this area, would probably extend it a bit further than that. He would likely say that once we understand it well enough, we can make anything we want out of biology, which I think is an interesting and provocative idea. I'm still just thinking about making actuators.
For a long time, I used to say that my vision, and I would hope it happens in my lifetime, is to make essentially a moving bag of meat; that we design a little skeleton, it doesn't even have to be biology material, it can be a polymer, and we seed it with cells, and we already know how to make those cells do what they know to do, and they find the right place, they grow in the right way, they attach in the right place. Take it out of the incubator, put in a little tiny microchip, and it crawls away. I think that's actually feasible, but having started to work on that ourselves, I realize that we're still a long way from being able to do that.
I would like to, though, make what I think is a really important point in the biological robotics realm. This is my personal view, and it's probably something we can debate. I don't think we should ever be tempted to make robots that can reproduce. I think that would lead to all sorts of ethical and physical problems. I think we need to be in control of how many robots we have, and where they are. In the factory, we can make millions of them cheaply, and they grow, but we don't want each one of those robots to be able to make another version of itself.
That might not be good. The other part that, even as a neurobiologist, I want to warn everyone about is, we shouldn't grow brains. I don't think that's a good idea. Why? First, I don't think we need them. Brains are pretty darn bad at doing neural processing compared with a microprocessor. People don't believe me, but they are. They work very well because we have 10 billion neurons in a brain. But any individual part of that brain, it's a pretty lousy processing unit compared with modern electronics. It can only fire action potentials, pieces of information, at about 100 Hz.
When you have a gigahertz microprocessor, we can't keep up with that. Our nervous systems can't do that. So as long as the job of the robot is relatively straightforward, a microcontroller, microprocessor is going to be much better than a living tiny brain. The second part is ethics. I think that the moment you give something a nervous system, it becomes an entity, an agent that we have to think about in terms of other living creatures.
Presumably this will happen, and we will get there, and we'll have to deal with it anyway, but I would want people to think about it ahead of time. We're already at the point where the octopus is receiving protection as a creature that has agency, and that we think probably has some, dare I say, sentience probably. If that's the case and we start to create artificial versions of those, we need to be very, very thoughtful about why we're doing that, and whether we need to do that.
We're a little way off from that being a problem in our field of robotics. But the more adaptable our machines, the more capable they are, the more we're going to have to deal with issues of agency, sentience, does this machine have rights? And it's not just about whether it can feel pain, it's to do with whether it has become so much a part of the society that to take away its rights would be an immoral thing to do. And it's getting into an area of philosophy that I am not qualified to talk about.
Another important thing is to make sure that the students feel they have the opportunity to question. These days, the role of a teacher, and it probably has always been true, but particularly these days, the role of a teacher is not just to convey information. The information's available to everyone. My students can access everything I know. The problem is, it's not organized, so I'm trying to organize it for them, but it's not just conveying the information. It's conveying a way of thinking in a critical manner, and to take what you're teaching people and have them be think critically about it, appraising it, and saying, “Is that true? Why? I don't think that can possibly be true because so-and-so...” Questioning all the time. The students need to feel that they can question.
What you're doing is triggering people's minds to turn. You're not just dumping things into a sponge, you're inspiring thought. Part of that is getting them excited. I think soft robotics is awesome for getting students excited because it's accessible. People can do some of this in their kitchens. Once you have them hooked into thinking about it, now you teach them the hard stuff as well, and they want to do it because it's interesting too.