A civil engineer explains how new insights gleaned from the flight of birds may one day be applied to fields as far-ranging as autonomous cars and crowd control.
Anyone who has ever observed a large flock of starlings in flight — darting and swirling as if the entire flock were one big beautiful being — cannot help but marvel and wonder at how all those birds keep from crashing into one another.
Nick Ouellette is studying the in-flight behavior of birds to draw lessons he can apply to engineering. He says that birds are not alone in their tightly coordinated patterns of movement; such behaviors can be observed at every scale of nature, from bacteria to bees to beluga whales.
Ouellette is doing sophisticated video measurements of flocks in flight to understand just how it is that birds can pull off their beautiful balletics without total chaos. He says the secret is that nature favors decentralized, bottom-up control of groups versus the top-down, leader-follower approach favored by humans.
Ouellette, a civil engineer and birdwatcher extraordinaire, discusses his research on the latest episode of Stanford Engineering’s “The Future of Everything” podcast with bioengineer and host Russ Altman.
Transcript
Russ Altman: Today on “The Future of Everything,” the future of flocking and swarming.
Now, anyone who has ever seen a flock of birds in flight has likely been impressed with the beauty and the coordination of the birds as they move, they turn in the air, they do these maneuvers that are just quite spectacular.
And we all look at this and I think we all say, How does that happen? How are they dividing and then rejoining, how are they making turns? Is there a head bird, as it were. Is there a head bird making all the calls, or did they in some way watch all the birds around them and take cues from them, is there a plan, or is this totally spontaneous.
Now I have always assumed that the positions of the birds are determined at least in part by what I would think of as optimal aerodynamics. If you’ve ever been on a bicycle, you know that if you get behind somebody who’s pushing the air out of the way, it is a lot easier to cycle than if you’re having to be the one in the front. And so I assume that the birds position themselves in a way that is in some sense aerodynamically optimal.
I actually have no idea if that’s true but I’m hoping to learn about that. Do the birds take turns at the front, in order to distribute the responsibility both for leadership, like for where we’re going, but also energy expenditure. I think the same questions might apply to schools of fish or swarms of insects. I think the details are probably very different, but the sense of wonder in watching them and seeing this amazing coordination of individuals is similar. In addition to the beauty of this phenomena, they also have practical utility. For example, will self-driving cars in the future fly in formation? Will they look more like birds, fish, insects, or maybe a herd of horses?
Nick Ouellette is a professor of civil and environmental engineering at Stanford University. Nick studies the physics, energetics and structure of flocks and swarms, and related phenomena. Nick, why would a civil engineer study these phenomena? We might expect it to occur in a biology department. And what have you learned that may be useful in a context of engineering new systems?
Nick Ouellette: Yeah, you know, I get that question a lot. The fully honest answer
Russ Altman: This is what we’re searching for on “The Future of Everything.”
Nick Ouellette: I study these things because they’re beautiful. Because they’re ubiquitous, you see them all the time. And it’s one of — A lot of what motivates my research is, you go outside, you look around you, you see something that’s interesting, you think to yourself, Gee, someone probably knows why this happens. And more often than not, when you do a little bit of reading you find out that that’s not the case. That you only have to scratch the surface before you get to some kinds of what seems like basic questions, where nobody really knows the answer. So that’s the real, full, honest answer as to why I’m interested in these kinds of phenomena.
Russ Altman: And your colleagues in engineering are OK with it?
Nick Ouellette: My colleagues in engineering are wonderful. And in part, I think that’s because it’s very very difficult in general to predict you know, twenty years down the line, what basic research now is going to lead to something really interesting in the future.
That being said, I think there are compelling reasons why one might want to look at these kinds of phenomena from the standpoint of engineering. And it’s part of a broader move over the past few decades toward bio-inspired engineering more generally, that you look to the natural world to see how evolution has solved problems, and if the same or similar problems show up in whatever you want to design, at least it’s reasonable to ask the question as how nature has done this.
Russ Altman: That’s totally fair. Let me ask you, when you look at these flocks in their beautiful formations, what are the problems you see them solving?
Nick Ouellette: And that is actually in some ways the most difficult question that we’ve found to address in all of this work. Because at the end of the day, you cannot go and ask the birds “So why you doin’ this? What’s the purpose?” You can make fairly reasonable arguments based on intuition from ecology, so interfacing with ecologists and biologists is really important here.
In some contexts — so I should say, I should back up just a smidge and say that one of the hints that collective behavior, as this broad field is known, may be really valuable from an engineering standpoint is that you see this at every single scale in biology, That individual units or organisms work together to solve problems. Everything from single celled organisms like bacteria, like cells, all the way up to whales. So you have literally the entire spectrum of biology that all cooperates to accomplish things.
Russ Altman: And there is an idea that many of these are distributed in the sense that there is not a central boss.
Nick Ouellette: Right.
Russ Altman: And then that leads to the question that how could you do that? Because so many human systems are hierarchal and do have a center but not all human systems are like that.
Nick Ouellette: But not all human systems. And in particular, not in the systems that evolution has designed. Humans are much like birds or fish or insects in that we have collective crowding behavior all the time.
In that especially kicks in when something is out of the ordinary enough that it shuts down our primary cognitive functions. If there’s an emergency, if there’s a disaster, and you’re in a big crowded situation, I don’t think any of us, if you think about it, you don’t make rational decisions. You don’t carefully weigh all the options — should I put my left foot here, should I put my right foot there, should I go in which direction. You follow the crowd to get away as quickly as possible — you think — from whatever is going on.
So there are reasons to think that the kind of — and this gets back to the question of why would a civil engineer want to do this — there’s good reason to think that things that we learn about how collective groups behave and how they can be manipulated may be able to valuably mapped onto questions about crowd control, about the design of public spaces to avoid instabilities of crowding behavior that leads to events like trampling or dangerous situations.
Russ Altman: Great, before we get to those engineering applications, I do want to go back to the birds, because I do really actually want to ask you the questions to these questions that I raised. In fact, is there a head bird, and how do they do the leadership of these flocks?
Nick Ouellette: The first thing to distinguish there — because the answer to everything is always this complicated — there are really two kinds of flocks you want to distinguish. And then we all have seen these.
Russ Altman: Good, we’re gonna have a taxonomy of flocks.
Nick Ouellette: Right, the first kind, which kind of looks like it has a leader though it probably doesn’t and where you may see aerodynamic benefits are the kind of V formations of flocks you see in large migratory birds. Geese, storks, ibis, all these kinds of things.
Russ Altman: Very beautiful V shaped with sub-Vs, like fractal with Vs.
Nick Ouellette: Yeah. That’s not what we have worked on so much. I know that literature anecdotally, but I haven’t worked on that myself.
Russ Altman: In general, are they aerodynamically pretty good?
Nick Ouellette: Yes, although it’s complicated. Everybody wants to make the analogy to bikers.
Russ Altman: Bikers, that’s one I did myself.
Nick Ouellette: Or trucks on the highway, trucks will do the same kind of thing. In fact our dean — one of the most recent papers I published on that asked me that very question. With cycling with as a you know kind of touchstone.
The difference really comes into the fact about how the birds are moving. And what’s easy to forget is that birds don’t just keep their wings fixed and glide, they flap. And the way aerodynamic drag works on a flapping wing as opposed to a fixed wing is very different. You can get aerodynamic savings, in fact it’s been shown that these birds do get aerodynamic savings. But the key thing there is actually the phase of the wing beats down the V chain. You have to catch the updraft
Russ Altman: Oh, that’s a beautiful thing.
Nick Ouellette: From the bird in front of you and then you get a boost from it. But if you’re flapping at the wrong time then you want to go up and the vortex shed from the bird in front of you is going down and then it’s bad for you.
Russ Altman: Right. Then this gonna be slightly different principles than the Blue Angels use because they’re all fixed wing.
Nick Ouellette: Right. They’re all fixed wing, and so drag works differently.
Russ Altman: This is “The Future of Everything.” I’m Russ Altman. I’m speaking with Nick Ouellette about — Well, the bird taxonomy we just learned about Vs, but there’s another important type of flock.
Nick Ouellette: Now think about the last time you went to a city and saw a bunch of pigeons flying around. Pigeons will do this, most birds, songbirds will do this, starlings are the classic example that form massive flocks. In those kinds of situations, the flock itself is not as rigidly sort of constructed as a V. The bird position is sort of more random. What we and others have found in those kinds of flocks is that in fact it’s aerodynamically bad, in that the birds wind up expending more energy to participate in a flock than they would if they were flying on their own.
Russ Altman: And they do look somewhat chaotic, just on a visual, even though they’re beautiful and coordinated, they seem to be making tighter turns and vortices and things like that that look like they might in fact be more energetically expensive.
Nick Ouellette: Yeah, and that is borne out by the data as well.
Russ Altman: And those are the ones that you study.
Nick Ouellette: Those are the ones that we study.
Russ Altman: Paint me a picture now for how the laboratory looked. What does it look like? What are the key tools that you guys use to make these studies?
Nick Ouellette: The first thing is —
Russ Altman: It must be cool technology.
Nick Ouellette: Yeah. And fact, the technology is how I got into this in the first place. The first thing to note in this front is that it’s not in the laboratory. Stanford is a wealthy university. They give you nice lab space, all these kinds of things when you come. They don’t give you enough lab space to have a few thousand birds Flying around in free space. This is very much a field project. I have a wonderful collaborator in England in Cornwall who has been studying the native jackdaw populations. The jackdaw are a small crow species, very social.
Russ Altman: And I’ve heard that crows are very smart.
Nick Ouellette: Crows are VERY smart!
Russ Altman: Which might confound your science. But we’ll get to that later.
Nick Ouellette: People are nominally smart too, right, so that’s —
Russ Altman: Well put!
Nick Ouellette: And that’s actually an interesting thing to ask about how does intelligence map on to this ubiquitous biological phenomenon, where animals and organisms that are not very smart do things that look similar to animals that we consider to be very smart. This is not a project that is done in a laboratory, it’s a project that’s done outside. That’s part of the reason why we haven’t focused on the V formation flocks. The technology that we use here is in principle fairly straightforward. We just use camera imaging to watch where the birds are going. The simplest kind of thing you would do is you go and take a camera, you point it at the sky, you see a bird flock flying around and you take some pictures of it.
Russ Altman: And it would be a movie.
Nick Ouellette: It’d be a movie.
Russ Altman: But that’s 2-D.
Nick Ouellette: That’s 2-D. And there’s only so much you can tell from a two-dimensional projection of what the birds are doing. What we actually want to do is to measure them in 3-D. We want to be able to locate in three-dimensional space where all the birds are, how they’re moving, and as much other detail as we can get. The way we accomplish this is similar to the way we see in three dimensions all the time which is to use more than one imaging device. We have two eyes, and that allows us, with some neural processing, to reconstruct a picture of the 3-D world. We do the same thing with multiple cameras.
Russ Altman: These flocks, I presume they are absolutely freely behaving, behaving, there’s no training.
Nick Ouellette: There’s no training.
Russ Altman: And you don’t physically even limit them?
Nick Ouellette: Nope.
Russ Altman: It’s catch as catch can in some sense.
Nick Ouellette: Yes, and this is why having a collaborator who has spent decades mapping out the behavior of these birds is vital to this project. Because the first thing you learn when you go to do field work to try to look at the sky to see birds is that there’s a lot of sky, and birds are very small. If you just randomly pick a spot and put some cameras out the chances of seeing a good bird flock flying over that is zero.
Russ Altman: This is “The Future of Everything.” I’m Russ Altman, I’m speaking with Nick Ouellette. Now about studying birds in the wild with 3-D capabilities. What do we learn? I mean I’m sure people have been dying to hear. Where is the science now in terms of understanding the coordination and distributed decision making of these flocks?
Nick Ouellette: There’s a number of things that people have learned about this. The first large-scale experiments, quantitative of measuring birds were with starling flocks from a group in Rome. They found a number of very interesting things.
There’s been a supposition in the field for a long time that it’s reasonable to expect that if you’re a bird in a flock, and it’s a dense flock, and there’s thousands of birds, there’s no way that you know what everybody else is doing. There’s not a chance. Which means that all of the coordination that you see has to come out of the decisions you as an individual make based on information from your local neighborhood. But one of the questions is what defines that neighborhood? Is it that there’s a distance in space that you can see? 5 meters? You just look at everyone there?
Russ Altman: These are the questions that as I was writing my introduction I said I have no idea what the answer to these questions are.
Nick Ouellette: The other way that you could think of defining what is a neighborhood mean is that you pay attention to some number of other birds And it doesn’t really matter if they’re five meters, ten meters, two meters away from you, you pick, say, eight, and then you follow that motion. There’s good reason to think that that’s biologically more reasonable, because you don’t get then overloaded if the flock gets denser. And in fact with the starlings, that’s what was found.
Russ Altman: And how do you prove something like that?
Nick Ouellette: It’s a very difficult —
Russ Altman: So you have a model that, okay, the bird is either looking at one or two birds in front of him, or an assembly of birds around them. I guess that leads to specifically testable hypotheses?
Nick Ouellette: Yes! So the way we do this is to make the hypothesis that if reliably you are paying attention to other birds at some distance away from you, either whether that being a counting distance or a distance in space there’s probably gonna be some spots you’d like to position yourself relative to those others that are preferable. You want to be able to keep them in view, maybe you want the aerodynamics to be not terrible. There’s gonna be some statistical bias. What you do here to get an estimate of range you see how far away do I have to push that range away from me before the statistics look uniform in space.
Russ Altman: Yes. That’s good, and so then that allows you to conclude that there’s like a set of birds around, maybe not even the few closest, but they’re a little more complicated than that. You made a vague reference to this a moment ago. Is there also an awareness of the I guess the aerodynamic, the pressure, the pressure environment around the bird, like the aerodynamics, I’m sensing turbulence here, I’m sensing smooth air over there, or is there no evidence for that?
Nick Ouellette: Honestly we don’t know, because we can now, you gotta imagine that the birds when you got a camera on the ground, you’re looking at a flock in the air. It’s very far away from you. At this point we can measure what the birds are doing, but we don’t have a local measurement of the fluid flow around them.
My guess is that the answer is probably no. That they don’t pay that much attention to it on average because the statistics you see don’t seem to matter sort of where you are in the flock, if you’re at the front, if you’re at the back, at if you’re on the sides. The fluid mechanics environment will be different in all those cases.
Russ Altman: Yes, that makes sense.
Nick Ouellette: So if the aerodynamics is a player, it does not appear to be a primary player in this.
Russ Altman: I know that a part of your work involves simulation. You go back to the lab, to the regular lab, not the beautiful Cornwall, you know — I’m going into a reverie thinking about what it must be like out there.
Nick Ouellette: Well it’s December. That’s when they do what we want them to do.
Russ Altman: Oh, okay, I’m updating my reverie! But let me ask when you get back to the lab, what kind of simulations are you doing, and how does that inform what you do next time you’re in the field?
Nick Ouellette: The simulations are a lovely place to test hypothesis. You can build a little model that says okay suppose we think that a bird pays attention to seven or eight of its neighbors or it pays attention to everything in some distance. You can put those rules into a computer and simulate what you would get and then try to look in those two simulations where you know what you put in. You never know with the birds, you don’t know what they’re doing. But you know exactly what you put into the computer and then you can see what the output looks like.
Russ Altman: And if it matches the behaviors — so that actually — so this is kind of a random question. Has the cinema industry approached you about getting better fidelity CGI simulations of bird flocks?
Nick Ouellette: They haven’t approached me because this has been done in the cinema since the 80s.
Russ Altman: Oh, so they’re pretty clueful about birds?
Nick Ouellette: Mmhmm. Actually one of the real seminal papers in this field was written for the computer science SIGRAFF conference in 1987 by a wonderful person who she actually lives in Mountain View just down the road. It’s a classic paper and it’s been used in actual feature films since the early 90s if not before, and in the video industry as well.
Russ Altman: Now are those things of interest to you, or are they in some way cheating in a way that it doesn’t give you the insight that you’d like?
Nick Ouellette: It’s not cheating, but I will say that we haven’t focused on it too much. Because one of the things that you start to learn when you do lots of different simulations is that many, many different sets of rules give you a visual output that looks very reasonable.
Russ Altman: They may have set upon something that gives you a perfectly good looking movie, but not related to the actual mechanisms used by the birds.
Nick Ouellette: Right. And this gets back to the engineering context in which I’m working. If at the end of the day I have learned what the birds are doing and then potentially apply this to an engineering problem and say “Evolution means that this is probably good,” if I understand what it’s being optimized for, I need to be pretty sure that I got it right.
Russ Altman: A perfect segue to our next segment. This is “The Future of Everything.” I’m Russ Altman. More with Nick Ouellette about birds, but now the engineering implications of all these learnings next on Sirius XM.
Welcome back to “The Future of Everything.” I’m Russ Altman, I’m speaking with Nick Ouellette about swarming, flocking. We’ve been talking about the study of birds and their flocking capabilities but we promised to get back to the engineering applications of these ideas. Nick, paint a picture for me about how some of these learnings are either directly useful, or leading to useful information.
Nick Ouellette: There’s a couple — I should preface this with I haven’t — we’re still sort of in the discovery phase of this kind of research. I haven’t personally spent a lot of time on the applications side. But two areas that I think are probably the most promising to apply this kind of work in engineering. One is on the question of human crowds, either crowd control or the design of spaces to manipulate crowds in a passive way. And on the control of distributed systems more generally. If you take the second of those for example, more and more in —
Russ Altman: And define distributed systems for people who might not think about them every day.
Nick Ouellette: In some sense a traditional way of doing engineering if you think about how you control something you build a system that’s sort of one big monolithic thing. Take the powering, one of the other things that we work on in the lab is questions about how you might want to design next generation power systems or electricity systems for cities.
In the standard old design is you build one big power plant and it pipes all the electricity to the city and you’re fine. Well these days, with the development of lots of local renewables, you could do that still, or you could put a solar panel on everybody’s roof, or you could do anything in between. Those would be examples of distributed systems, where you have many, many, many components to the system that all need to interact. But as an engineer you still need to have that system be controlled. It should do what you want it to do, it shouldn’t fail, it should be robust, all these kinds of nice things.
Traditional engineering control work is a very top-down kind of thing: you want to know what is the total information in the system, what is everything doing, and then make decisions based on that.
As we were just arguing, the birds or animals don’t do that. A single bird doesn’t know what the other birds are doing. And yet the flock itself is very coherent and very consistent in its behavior. This suggests that there’s an alternative paradigm where you use only local information with local interactions that are of the right type such that the entire system still has functionality and is very very robust.
Russ Altman: Got you.
Nick Ouellette: And I would argue that that kind of bottom-up control scheme should be at least be on the table. As engineers we move more and more toward thinking about distributed systems rather than monolithic systems.
Russ Altman: And are you thinking about distributed systems like self-driving cars and trucks?
Nick Ouellette: Absolutely! or from your perspective, are they all about the same? Well, that’s a good question. I would say there are aspects to all of these these things that are about the same. Once you — and this goes back to your question about what do you learn about putting on the computer in a model.
In some sense, once you have abstracted the real system into a model and you’ve validated the model, well now it’s just math. And anything you can apply that math to, so any different mapping, you have a variable in your model and you know it came from meeting a bird, and now it means a car, doesn’t really matter from the math standpoint.
Russ Altman: Now, but let me ask, because you’ve mentioned two things have come up in our conversation: Humans and smart crows. And they have social structures. Does that matter?
Nick Ouellette: That I think is actually a very very interesting thing. And this is where a lot of the recent work we’ve been doing with animals where we’ve been focusing. The answer to that question even a couple of years ago was not known. You can definitely say that humans have certain social interactions. Does that matter in a crowd situation?
One of the reasons I was excited to work on jackdaws is that unlike what we think about starlings — the biology is always difficult — we know that jackdaws and other crow species have very intricate social systems. And for jackdaws in particular we know that they mate for life. You have a society that’s built of these mated pairs. In the roosting season, in the winter months in England where there’s no young in the nest, both members of that pair are free to fly around all day. When they form these flocks, anecdotally people have seen these and said it look like that flock is composed of lots of little pairs of birds.
Russ Altman: And you can see the pair behavior in the flock?
Nick Ouellette: You can see the pair behavior in the flock. And we can pick that out automatically. Once we get our data set, we’ve imaged the flocks in 3d we know where all the birds are, you can actually pick out which ones are paired together.
Russ Altman: And it’s something like these two maintain a close distance over a long period of time —
Nick Ouellette: Over a long flight —
Russ Altman: Which allows us to infer that they are spouses. I don’t know if that’s the right word.
Nick Ouellette: Partners.
Russ Altman: Partners, thank you.
Nick Ouellette: And they also maintain similar configurations over time so that the placement of each bird relative to each other is also fairly consistent. There’s a lot of statistical measures we can throw at this to determine what those pairs are. And then you can start to ask the question, okay great you found pairs. Does it matter from the standpoint of the flock that there is structure on the inside?
And the answer, in some sense surprising answer to that is, yes it does! And what we found is that the more pairs you have in the flock as a percentage of the number of birds, because not all the birds —
Russ Altman: There are some bachelors —
Nick Ouellette: There’re some singles —
Russ Altman: Bachelorettes, whatever, forgive me everybody.
Nick Ouellette: The more pairs you have, the less responsive the whole flock is to perturbations from the outside world, things like predators.
Russ Altman: This “The Future of Everything.” This is Russ Altman, I’m speaking with Nick Ouellette, and we’ve just learned that the mating pairs in this case of these birds actually affects the dynamics of the flocks. So please!
Nick Ouellette: The more pairs you have, the overall flock is sort of not as effective at doing things like avoiding predators as it would be otherwise.
Russ Altman: It’s a little bit of a negative?
Nick Ouellette: Little bit of a negative, for the society, for the group. But at the individual level, that is balanced by a couple of interesting savings. What we’ve also found is, so I’ve told you okay the birds pay attention to some number of neighbors. Well that number is smaller if you’re part of a pair versus if you’re not.
Russ Altman: You’re focused on your kin.
Nick Ouellette: Well it’s typically about three, three or four. Definitely your partner, and then a couple of others close to you. But about half the range, half the number of birds you pay attention to if you’re not part of a pair. And that has an implication back to the aerodynamics.
What we found is that it’s always an energetic hit. There’s an energetic cost to being part of the flock. Which is reasonable to pay if you get benefits like a hawk comes by and eats one of your neighbors instead of you. But that cost is mitigated at a local level if you’re part of a pair. You’re not paying attention to as many birds, it’s more beneficial for you, so there’s an individual level savings but a global cost to the group.
Russ Altman: And it’s not hard for me to imagine similar dynamics in a crowd fleeing an emergency where your partner, you’re keeping an eye on them, you are trying to save then if they’re in a tough spot where otherwise nobody might try to save them, but that might lead to a less effective emptying of the theater.
Nick Ouellette: Or at least something you need to know about when you start to try to poke and prod at that and manipulate it.
The other area where I think this might be interesting to think if you think about self-driving car fleets and controlling the with these kinds of local rules Well you got your Google cars, and you got your Uber cars, and you got all these other kinds of things. And even if on paper, they’re supposed to be communicating in the same way, protocols may be slightly different, there may be asymmetries.
Russ Altman: Right
Nick Ouellette: And what we’ve been finding in the birds is that those little local asymmetries can have consequences.
Russ Altman: Now we have a potential — I’m not sure it’s an engineering principle yet — but you’re pushing close to a principle that we may want to evaluate the costs and benefits of having different algorithms on the different cars, because that may lead to a similar behavior as birds that are caring more about some birds than other birds.
Nick Ouellette: Or at least it’s something that one should be aware of As we move towards using these kinds of rules that we’re learning from nature and applying them in engineering contexts.
Russ Altman: When will these principles affect people in the real world? When are we going — and I know it always takes time.
Nick Ouellette: Always takes time.
Russ Altman: Give me a sequence of events that might lead to these principles affecting the systems that are being built.
Nick Ouellette: That’s a question I probably can’t answer. I don’t know that I want to speculate too much.
Russ Altman: Please do, we’re in the last thirty seconds.
Nick Ouellette: Yeah, if I had to guess, I would guess that it’s gonna be in the crowd control arena that you’re first going to see things like this. ’Cause that’s being, that’s a longer, because we’ve known about crowds a lot longer than self-driving cars.
Russ Altman: So human crowds.
Nick Ouellette: Human crowds, I would guess that there’s gonna be-
Russ Altman: And perhaps the way that they set up exits, the way that they engineer buildings for entrance and exit.
Nick Ouellette: This is already something people are thinking about as we get more data, I think we’ll things happening there.
Russ Altman: And that’s actually exciting for improved public safety.
Nick Ouellette: Absolutely.
Russ Altman: There are cities, the cities are crowded. Fantastic. Thanks for listening to “The Future of Everything.” I’m Russ Altman. If you missed any of this episode, listen any time on demand with the Sirius XM app.
Source: engineering.stanford
