A Robot Fly at Harvard
Interview with Robert Wood
ROBOTS (Sabine): Hi Rob. Can you maybe introduce yourself to our listeners?
Robert Wood: Sure, my name is Rob Wood. I am an assistant professor of electrical engineering at the School of Engineering and Applied Sciences at Harvard University, and I am the founder and Director of the Harvard Microrobotics Lab.
ROBOTS: What are the main research goals of your lab?
Robert Wood: One of the primary research goals is to use biology as inspiration and extract underlying principles that give certain species remarkable agility, maneuverability and robustness, and use those principles along with the best engineering practices and techniques to recreate robots which mimic those species. For example, one of our primary projects is the creation of a flying robotic insect approximately the size of a house fly.
ROBOTS: So you have not been looking into [just] any type of robots; you’ ve been looking into the small robots of [?] centimeter [to] micro-scale robots?
Robert Wood: Right. Perhaps it is best to start off with the motivations that we have for what we are doing. If you imagine a small autonomous robot that is extremely agile, the applications are as follows: you could put these things in places where you would not want to put a human or an animal. For example in a collapsed building in a disaster situation where perhaps an earthquake has collapsed a building, and you want to search around the rubble and try to find survivors, well any of the current autonomous robotics platforms are going to get stuck at very moderate rubble, whereas if you have something which is the robustness, the agility, the size of an insect then you can perhaps do search and rescue tasks very quickly, very cheaply, and robust in the sense that you have potentially many thousands of these agents swarming around such a site, and if even 99% of them fail and one of them finds a survivor, that’s a success.
ROBOTS: What type of insect robots have you been creating?
Robert Wood: As I said, one of the primary projects is the creation of a robotic fly, and I could argue that that might be one of the biggest challenges is to recreate flight at the scale of an insect, for this sort of class of robotic insects. That is based upon some of our new techniques in this sort of mezzo scale of fabrication regime, and based upon these fabrication techniques that we have developed, we are also using those and leveraging those for other classes of robotic insects, things that crawl and things that swim as well.
ROBOTS: Let’s concentrate on the fly robot. What does it look like? What does it sound like? How does it fly?
Robert Wood: Sure. Well, it looks and acts remarkably similar, at least mechanically remarkably similar to an actual fly and that is not a coincidence. As I said, we extract principles from biology and try to understand them and then apply them, with bio-inspiration as our starting point. We have used some of the work of some of our biologist colleagues, for example Michael Dickinson who is now at Caltech, who have given a really nice qualitative description of the complex aero-dynamics and thoracic mechanics of actual insect flight and how they are able to achieve the astonishing maneuverability that they achieve. And so when we recreate these structures, we work directly off of that blueprint and for everything from the wings to our mechanical drive system, even our actuators are in some ways similar to biological muscle, and so it is not a coincidence that the structure actually looks like an insect, like a fly. It has two wings and those wings beat, approximately, greater than a hundred times a second, which is consistent with I guess you would say a fat housefly. [It] weighs about the same as a fly, is about the same size in terms of wing span as a fly, and moves its wings nearly identically to how a fly would move its wings, while it is hovering, and the wings themselves actually are a close replication of at least the shape and the structural properties of a fly’s wings.
ROBOTS: What has this fly robot been able to do so far?
Robert Wood: So we started this project with the goal of creating a fully autonomous system, and we thought that the best way to approach this — and this actually started in Berkley in Ron Fearing’s lab, when I was a grad student there, and has continued into my lab at Harvard — the way that we wanted to start this was by actually figuring out a way to create the structures necessary to produce enough thrust to get off the ground, and so that consisted of coming up with the designs as well as the fabrication paradigms to actually make the mechanisms at this mezzo scale. I am calling it a mezzo scale because it is basically too small for the traditional macro scale machining and has requirements which exceed what could be typically accomplished in MEMS, so I am calling it a mezzo-scale manufacturing paradigm that we have accomplished. And so we wanted to come up with the design and the fabrication methods to be able to recreate both the articulated structures, the actuators necessary to power the fly, the wings, the exoskeleton, all the different components which make up the mechanics and aeromechanics of the fly. That was our first goal, to show that we could actually recreate the insect wing motions, at the scale of an insect, with the same sort of kinematic patterns as an insect, and the second part of that goal was to say well do those wing motions, at this scale, do they produce the thrusts that we expect based upon what we understand of insect flight and is everything consistent? And so what our flies are currently able to do is reproduce the motions of a fly very well, in terms of what a fly is doing in hover. And then we have taken that integrated fly, put it on a force sensor, and measured the forces, and it is also consistent with what a fly would see and then actually allowed it to take off, and when I say take off, I use this term a little loosely in the sense that it still has external power, it’s tether for external power, and it has constrained body degrees of freedom, but indeed the integrated package was able to lift itself off the ground, with those caveats, so that is what we were able to and then now, and, of course, due to the fact that it is externally powered and uncontrolled, that leads to a great number of research topics which we are having to address. Now comes the fun part, now that we have proven that we can actually make these structures on that scale, now it is how do we control them, what are the electronics that are going to go into them, what are the sensors that are going to go into them, and how we are going to power them?
ROBOTS: Can you give us some insight on what this power and high level control maybe will be?
Robert Wood: Absolutely, sure. Power I consider to be both the easiest and the hardest question simultaneously, easy in the sense that we know that thick film batteries exist with the energy densities sufficient for a decent flight duration and power density sufficient for energetic excursions, and based upon what we know are quoted as energy densities and power densities and the sizes of batteries currently available, along with what our power draw is from that we can actually measure from our robotic insect, we predict about 5 to 10 minutes of flight time. So that is the easy part in the sense that we can sort of take these things, not quite off the shelf at this point, they are still in the research phase, but we are confident that we can use those. Now, the hard part of the power question is if you want to extend it beyond that 5 to 10 minutes of course, and we think we can accomplish that in a couple of different ways. First would be to increase the propulsive efficiency of the mechanical and aeromechanical structure and that is something that we are very focused on in my lab. The structure that we have built is in no way optimal in term of mechanics or aeromechanics. The second way is to increase the energy density of batteries. That is something we are not particularly interested in but it is happening with and being driven by other factors, not as fast as something like Moore’s law but it is happening. And then the third thing that you can potentially do is use some energy scavenging, energy harvesting techniques such as solar or vibrational, and if you use some of those perhaps you are not going to be able to extract enough power to power flight but perhaps you can fly around, expend your battery, find somewhere to perch, once your battery is expended, and then recharge yourself, and perhaps your flight duty-cycle will maybe only be like 10%, but you effectively extend your lifetime indefinitely. So that is the power question. And then the next big component is sensing electronics and control. Electronics has a number of different components to it. First and sort of the most obvious one is that well our actuators are field driven actuators meaning that they have relatively high fields, relatively high voltages, and if we assume we have some sort of battery voltage then our dry voltages are on the order of a couple hundred volts, so the first component of electronics is how do we get let us say 3 volts to 300 volts, and do it in a let’s say in a less than 10 milligram package, and do it with greater than 80% efficiency, and it is obviously not a trivial thing to do. Our lab has actually recently demonstrated 3 different circuit topologies that we think can do exactly that, and we are well on our way to achieving that goal. The second component of the electronics is the drive electronics. Once you have those high voltages generated, how do you actually efficiently drive the actuators in your structure and there [are] some methods that we have for that as well. The next component of electronic sensors, which we can split into 2 classes: sensors for control, for stabilization of the structure, and then sensors for navigation. In terms of the sensors for control, we have actually in the past built a biologically inspired sensor suite for attitude estimation, for use in body stabilization. That is based upon a couple of sensors that flies actually use in flight both photo receptive, mechanoreceptive sensors which give things like body orientation, body angular velocities, and do this in ways which are very power efficient, very small. We have a couple of prototypes; we have not integrated them onto a fly yet, but we have a couple of prototypes which show that a proof of concept for some of those. And some of them are relatively trivial, some of them are rather complex, and that is something that we are also very focused on is the development of any new sensor technology which will give us any sort of useful state information. And then of course the last component is once you have all this state information, is what do you do with it and how do you use that to control a fly? Now, one of the key things about the flying insect that we are trying to do is that it is very unstable, and that is something that we like. Typically, you want to make a robot which is either passively stable let us say or at least easily stabilizable, and in fact we do not want to do either. We want it to be very unstable, because we want to exploit this dichotomy between stability and maneuverability. We want to achieve [this] really nice, remarkable maneuverability that a fly can achieve, and a fly itself is very unstable, and so the starting point is that we have a very unstable system, and it is going to typically be quite under-actuated, even without considering the infinite dimensionality of air, and it is going to not have very linear dynamics, all these sort of classical things which make for a rather difficult control problem, but it is nice in the sense that it is also a very tangible control problem and something that we can get students excited about. So you could go about it, I see it in one of two ways, to do control. The first way, sort of the classical way is to come up with as good of a model for the nonlinear dynamics as you possibly can, considering all the things that I mentioned about nonlinearity and under-actuated, and see if you can come up with a control law or a control technique which will stabilize that system about some equilibrium or let us say hover for example. And there are certainly some techniques which would be applicable to that, switching controllers, etcetera, but you have the additional caveat of course that you are going to be processor limited, meaning you are not going to have an infinite amount of computation on board. You are certainly not going to be able to carry around a Pentium, so if you can even come up with a control law which is going to stabilize the system, is it even going to be feasible to implement that, and so that is one of the techniques, that is one of the paths that we are interested in, but it is not really as interesting per se as some of the bio inspired control approaches. So the second path, which as I mentioned is more interesting to us, is thinking about how to use some of the biological control architectures that are present in a fly, and adapt them and use some of those similar techniques to control our flies. Insects, flies I should say, have some really interesting control properties, and I am not a biologist so I might butcher this a little bit, but basically from my understanding is that the neural pulse trains which control the flight muscles which effectively control the wing kinematics of the fly, which effectively control the body torques on the fly, those neural impulses to those muscles are generated not by a central nervous system but instead by sensors, by small mechanoreceptive sensors, called the halteres, which are sensing angular velocity, so they are little gyroscopes, and these gyroscopes output control signals to effectively create a riding response to any perceived angular velocity, and so this concept of this local sensory-motor control loop, which is very efficient, very fast, and very computationally trivial almost, in the sense that you do not need to do much computation on the output of the sensor. And then the haltere itself, that little control loop can be suppressed by let us say a higher level of centralized controller which could then trick the haltere into thinking that it has a falsely perceived angular velocity, which then would trick the haltere into giving a riding response which is actually going to be a divergence, and cause the system to go into instability, which is going to cause a rapid maneuver. So these are the two paths and obviously to us the more interesting one is the bio-inspired approach, and they are two of the hottest topics, and this is collectively one of the hottest topics in my lab now.
ROBOTS: You have been taking a lot from biology. Do you think you can give something back?
Robert Wood: Sure. It is a hope of mine, has not really been realized yet, but one of the things that we can do is come up with relatively faithful recreations of some of these bio-mechanical structures; wings, driving the wings through various trajectories, and the actual mechanics of the thorax. We can do these and recreate some of the structures that are seen in nature with some degree of faithfulness in terms of what is actually in nature. And so if you can do that then you can potentially start answering some of the underlying questions in biology by using the robotic mechanism to explore some of the design space that surrounds some of these biological systems. So I will give a more concrete example. All insect wings create some degree of anisotropic compliance, and it is currently unknown why this arises, so what I mean by anisotropic compliance is that they are typically stiffer in the longitudinal, the length direction, than they are in the shorter direction. And so why is this? Is it some benefit in terms of propulsive efficiency and maneuverability, or is it just some bio-material limitation, meaning that the insect wings are limited to this structure of veins and a membrane, and the veins are typically something like chitin, and they are limited in terms of material, so that is currently unknown, but we can create wings with any sort of stiffness, and we can use things like carbon and exceed the stiffnesses of an insect, or we can make it more compliant, or anything in that range. We can exceed that design space, with respect to what insects actually have, and look at things like shape, inertial distribution, texture, stiffnesses, and then create a subset of those wings, drive those wings to any trajectory that we want really, and then directly measure things like peak thrusts, propulsive efficiency, and answer some of these questions. So to say what is the actual effect of this, let’s say, compliance on the performance of the wing and the wing drive, and then perhaps, hopefully that is interesting to some people who are interested in bio-mechanics. So that is one example where I consider this to be sort of a closed loop biological inspiration, where we started with trying to recreate a fly and now we are hopefully using some of the techniques that we use to actually build flies to then give some insight back into those systems.
ROBOTS: What area of robotics do you think will have had the biggest impact on our lives?
Robert Wood: I think that this whole emerging area of bio-inspiration has a huge impact on robotics, which in turn will have a huge impact on our lives I guess. But in terms of impact on robotics, I think that the concept of using things in biology and understanding them and just using that basically as shortcut to potentially enormous design spaces where if you want to achieve some goal, well if it is already been achieved in nature then why not start from that and work from there. I see this as really accelerating the pace of engineering as a whole. There are a lot of instances of this of course outside of robotics and engineering and sciences but I think it has really helped out in robotics, this concept of bio- inspiration
ROBOTS: Last question. How do you see the future? Can you imagine millions of these insects flying around, maybe helping out in different tasks?
Robert Wood: Well sure. The applications that I mentioned of search and rescue, or you can expand upon that and say maybe environmental monitoring or traffic monitoring or making really cool toys. Whatever it is, I see us hopefully having a autonomous hovering of one of these devices in our lab in the order of 5 years and maybe for more general use, maybe 5 years after that. And yes, I think that it could potentially be quite impactful in a positive way on a lot of the different things that we do, whether it is civil service type applications, or if it is entertainment, or a couple of different things you can think of, yes I can see not just flying but things that crawl or things that swim could do very useful things for us, and be quite interesting both scientifically and sociologically.
ROBOTS: Thanks, Rob, for sharing your work on microrobotics.
Robert Wood: Sure. My pleasure.
A Robot Fly at the MoMA
Interview with Paola Antonelli
Rob’s robotic fly then took off to the last floor of the MoMA, where it was showcased from February to May, in a futuristic exhibition mixing science and art. Adam has more on that from the Museum of Modern Art.
ROBOTS (Adam): Good morning, Paola Antonelli, and welcome to robots.
Paola Antonelli: Good morning, Adam.
ROBOTS: You were the organizer of the Design and the Elastic Mind exhibition, at the Museum of Modern Art. Can you tell us a little bit about it?
Paola Antonelli: Sure. It’s an exhibition that was focused on the relationship between design and science, and based on the principle that designers are the best interpreters of technological and scientific revolutions. When scientific revolutions happen, what designers do is they try to absorb them, and metabolize them, and then transform them into objects that everybody can use. So what they do is they bring revolutions into people’s life.
ROBOTS: And how did robotics play in this exhibition?
Paola Antonelli: Robotics is extremely important for the relationship between design and science from many different viewpoints. On the one hand, there is the idea that robots will be able to help human beings, so it is a very human-centered view, and kind of self-centered view, that tries to explore how different tasks and different functions can be actually performed by robots. But, also there is another viewpoint, which is the viewpoint that I think you are most interested in, which is robots that are studied as exploration of how nature already moves and of how nature already builds, so as to achieve more economical and more elegant ways to incorporate movement, dynamics, and also construction methods in our lives. So both aspects were presented in the show, the robots as helpers in a more ironic way, with some work by designers in London that are postulating robots that will not take care of us unless we take care of them first, very needy robots. And instead the part that is more scientific, where Doctor Woods’ work also falls, is the part of the exploration of movement.
ROBOTS: Can you tell us a bit more how Professor Woods’ work fits into this because to us engineers and scientists? It is a big feat of engineering, but I am wondering how this looks to a designer.
Paola Antonelli: To a designer it looks like bringing everything back to the basics, and starting from scratch in order to understand a particular method of movement, or a particular method of construction. So it is something that some designers really need to learn, how to really go back to the basics. In the exhibition, of course, we had Doctor Woods’ work and we also had a few prototypes by Festo, that were exploring the way certain birds fly and certain fish swim, plus we had some rehabilitation robots that were provided by Panasonic, for mimicking muscle. Pneumatic muscle seems to be one of the areas that many companies focus on right now because it presents immediate applications in rehabilitation. But even more speculative work, like Doctor Woods’ or also the exploration of the movement of centipedes, these kind of basic and very simple apparently but in reality extremely complex movements seem to be also something that designers can learn a lot from.
ROBOTS: So you are talking a lot about bio-inspiration, and how machines can move more like biological organisms.
Paola Antonelli: Absolutely, and therefore also reduce the consumption of energy, be more efficient, have a better return, and in fact Doctor Woods’ work, and the same with the Festo prototypes, were in one particular room of the show that was devoted to nature. And, of course, we have been studying how nature builds and how nature moves for centuries, at this point trying, trying to really understand the economics and the elegance of these secrets. And, paradoxically, one of the most artificial things we have ever constructed, which is the computer, is bringing us the closest to understanding nature and organic design and organic dynamics. So this particular room was devoted to biology, biomimicry, bioengineering, and yet it contained not only robotics but also experiments in computational design, for instance those that are conducted by Neri Oxman at MIT, in which she tries to look at natural phenomena and extract and distill algorithms from them, that will then enable architects and designers to build in a more thoughtful way.
ROBOTS: Did you by any chance look at commercial robots such as the Roomba vacuum cleaners or lawn mowers?
Paola Antonelli: Not this time for this particular show, because this show was truly not about today but about the day after tomorrow, so everything in the show was plausible and feasible, but not yet realized as a product. It was mostly concepts, speculative experiments, and very few products. But of course the area of robotics that already is part of our lives is of interest for MoMA. We do not have Roomba in the collection, per se, but it is something that we considered before and that we might acquire in the future
ROBOTS: Excellent. Well thank you very much for taking time to speak to us today
Paola Antonelli: You are welcome. It was a pleasure
ROBOTS: Okay, take care then. Thank you
Paola Antonelli: Take care too. Bye bye.