 My title from the original one that's written, but there's going to be a selection of ant-related work that I've worked on over the last, I guess, eight years, actually. And so let me start off with just thanking the folks who helped me with a lot of this work. So I've been at UC San Diego for about three years now. A good portion of this talk is going to be related to work that my postdoc, Glen Edd Clifton, has been doing there. And this is in collaboration with an ecologist at UC San Diego, David Hallway. Can everybody hear me okay, by the way? It's not too loud. Some of the work at the end where I'll talk about some of the robotic stuff that we do in my lab is done by a PhD student, Mingsang Zhang. And some of the work at the beginning of this talk I did during my PhD work with Dan Goldman, who I guess will be here later today. So let me just go over a brief outline. You know, I want to kind of go through all the things that fascinate me about ants. In particular, the way that they move individually, the way that they move collectively, the way that they construct underground environments. And I want to address some issues related to sub-training and locomotion, particularly how these environments are constructed, how ants move through them individually, how they move through them collectively. I want to move to, excuse me, then talking about sort of overground locomotion, so how ants forge walking across rough substrates. In particular, looking at the aspects of how substrate complexity and roughness affect locomotion. And then at the end, since we do have robotics in the title of this workshop, I'm going to talk a bit about the work that I do trying to create sort of insect-inspired robots. One aspect of that is how do we miniaturize robots down to the scales of insects themselves. Another aspect is how do we take principles of the insect exoskeleton and sort of body design and embed that in larger scale robots themselves. And so, you know, fundamentally I'm really, I'm in love with ants. I've worked with ants for over 10 years now. I think they're such a fascinating species. They exhibit beautiful collective behaviors, wonderful examples of highly capable locomotion and incredibly small packages. And I think in particular what fascinates me the most is that they live in an incredibly complex natural world. So the world to an ant is potentially far more complex than the world of a large scale vertebrate. And this is a concept that was originally proposed by Kasparian visor in a sort of hypothesis, the size grain hypothesis. That, you know, if you look at the savannah and you look at the animals that live there, larger scale animals like elephants basically see a nice flat world that they get to move around relatively freely. An animal such as a mouse or something like that might see a slightly more complicated world with some brush that has to contend with. But the animals that really living at this sort of millimeter scale see a very complicated, you know, kind of quote unquote fractal world that they have to contend with. They have to move through to forage to get from point A to point B. And so I think that locomotion at these scales, this sort of millimeter centimeter scale of insects is really fascinating, presents incredibly complex motor challenges for them to sort of successfully live in these environments. But they also have the capabilities of manipulating the environment themselves, which also presents fascinating opportunities for optimizing locomotion or making locomotion effective for their own body types. Ants are fantastic as well to study because there's huge morphological diversity among the ant species. So if you look at the sort of shape variation that we see across ant species, this is an example of two ants. And I do not remember the species names, but this is the largest of the ant that Kasparian visor were originally looking at for their size-grain hypothesis. This is the smaller species. This drawing right here is a comparison. This is the body size of the smaller species compared to the larger species of ant here. And in a more recent paper that was on the cover of JB, you can see this more vividly. This is looking at the, comparing the sort of vision capacities of ants across size. And this is basically comparing the smallest of the ant that was studied in that study of visual size sitting on top of the eye of a larger species of ant. So, you know, we see huge size variation among ant species. We can even see really large size variation within individual species within a colony of ants. Huge morphological variation just among different casts of ants. And, you know, not just in size, but in limb shape and limb design and head shape and all sorts of different morphological features. So I think ants present a wonderful opportunity to look at how sort of morphology compares to locomotion function and behavioral function. And we can look at these questions very specifically within species and across species. And so, you know, in this first part of the talk, I want to talk a bit about subterranean locomotion. So many ant species live below the surface. They construct subterranean nests. They do this by interacting with the soil, using their mandibles to manipulate soil particles, carry them up to the surface, deposit them at the surface. And, you know, one of my favorite videos is this right here. I didn't take it, unfortunately, but I'm well familiar with it. This is a fire ant nest and what it looks like when you're, you know, maybe living in Georgia and you go out and their fire ant nest basically every 20 feet around you out in a nice cleared field. You take a shovel, you crack into the surface of it and you basically get a spewing out of hundreds of thousands of these little fire ants. Fire ants are about three millimeters in body size. They live in huge colonies, so tens to hundreds of thousands. A fire ant nest, a casting of a nest by the biologist, Walter Chinkel, looks something like this. So maybe a meter below the surface. So you have huge populations living in these subterranean nests. They go basically, the tunnels go down vertically. As well as going down, they also go horizontally. And so this is a mapping of the foraging tunnels of two of these colonies, two such colonies of the fire ant. And let me see my scale bar here is, I believe, this is a 10 meter scale bar here. I believe the sort of total length of tunnels can exceed 50 meters in length. And again, these are about three millimeter individuals. So they create these huge foraging highways. These are fire ants. Yes. Did I not say that? I'm sorry. Yeah. So the first part of this talk is going to deal with fire ants. And again, this is work I was doing in Georgia Tech. So they're very easy to species to work with out there. Again, you step outside your door and you're immediately accidentally stepping into a fire ant nest. So I just think these forging tunnels are fantastic. And they create these huge subterranean environments. And so we were interested in basically all facets of this process. How do you create these environments? Who within the colony is digging or performing digging tasks within the, within a fire ant colony? This is a particularly interesting question of this, who is digging? Because if you look at the variation of the size of individuals within a fire ant colony, so this is a visual demonstration of the variation of body size within Solonopsis and Victor. And what you see is that from the worker cast, which is basically the only cast within a fire ant colony, there's about a threefold variation in the largest or from the largest to the smallest individual. And so one of the original hypotheses was that this, this size variation may lead to functional variation. So large workers may, you know, defend the nest more, more prevalently than smaller workers. They may perform more mechanical tasks such as digging or, or, you know, carrying away heavy payloads. So we tested this by isolating individuals, larger individuals and smaller individuals from fire ant colonies. And we basically challenged them to construct new tunnels within a freshly wetted sort of monodispersed soil sample, quasi 2D ant tunnel, just like you would have as a kid. So we did this in a large capacity where we measured eight tunnels at a time, sorry, eight farms at a time. We let this run for three days. We measured the final tunnel constructed area. So cross sectional area just as a proxy for how much total volume was excavated. And what we found was that basically if you look across the small cohort, the large cohort and a control group, which was just the natural population, there's no effective difference between the amount of excavated soil over three days. Independent of, you know, the fact that, the irrespective of the fact that across the small and large cohorts, there's about a three-fold difference in body size. So a three-fold difference in body length too, so just in terms of length scale, not in terms of volume and sort of muscle capacity and things like that. So, you know, that's one surprising aspect of at least working with fire ants is that across this continuous gradation of body size, there seems to be no specialization for digging tasks. Everybody seems to perform digging tasks at the same time. But the number of agents is the same in order? Correct, yes, yes, yes. So is it a question or is it that they are borrowing the most payload in over the 80s? So that's a good question. So you're sort of highlighting this down here, is that right? Yeah, I know. You start this video a billion times and you see that they tend to follow the more saturated region down here. Yes. So, you know, we looked at later on specifically how soil cohesion and sort of wetness properties affect digging, and there's a sweet spot in there in that you need to have some moisture present so that it stabilizes the grains. Too much becomes problematic in terms of forming the pellets that they carry to the surface. Specifically, you know, the kind of inhomogeneities that we see down here, you know, they're sort of frustrating in the sense that they do seem to affect the speed at which they dig here. But, you know, beyond that, I can't say anything more. I do think that they tend to follow sort of specific gradients of water. Is there any other questions? Yeah. Have you measured the excavation rate? We did. You're asking with respect to the gradients and whether they're following the gradients and speeding up or... As they move farther away from the surface, they take them longer to move the grains. Indeed. It does. It does. And, you know, you see basically a sewing down. It could be a square root of T type behavior. You know, a number of these, there was a, you know, I'm kind of speaking anecdotally now, but there is definitely a slowing down. So, you know, the slope is less than, or the slope is getting smaller. But you see interesting sort of behavioral bursts. So, you'll see it slow down and then it'll pick up again and then slow down. I can't, I don't know why those particularly happened. We did individual experiments where you force them to dig a single tunnel. So, you basically just make, put grains in a tunnel. And that, you do see a, you know, a slowing down. It's a process of having to walk an infinitesimally further distance as you go to pick up the next parcel. So, that's, it's well described by basically a constant velocity walking model where you're picking up individual grains and you're growing your tunnel by that process. So, next we looked at, you know, more specifically how individuals are interacting with the tunnel face and how they're transporting grains. And this is work that I collaborated with a postdoc on at Georgia Tech, kind of trying to map out the different specific behaviors that we see in terms of leg bracing, limbs are being used to manipulate the grain surface. You know, interestingly, and we'll see this again pop up. They use their antennae for formation of grains sometimes. So, you know, antennae are not just these sort of sensory appendages. They do seem to have purposes in a sort of more mechanical task. And again, we'll see that in a different context a little bit later. But in particular, you know, variants construct arrangements of grains and they carry them to the surface. And these arrangements of grains can be quite large. There is a limitation on that size because you've got to get it through the tunnel so the grain arrangement can't be bigger than the cross-sectional diameter of the tunnel. And so there's some interesting interplay between the size of grains and the manipulation of those grain balls. The next thing we looked at was traffic control. So looking at how individuals move through tunnels collectively, right? So you're living kind of in this collective environment where you have to get around each other. This happens, you know, in natural tunnels, typically in pure darkness, right? So you don't know whether there's a jam ahead of you or not. The other thing that we found was that there is a sort of traffic flow at a rate associated with tunnel diameters. So larger tunnel diameters had less propensity to form traffic clogs. But, you know, most interestingly, the ants sort of through behavior frustrate themselves, if you will, cause traffic jams, maybe unnecessarily because of this need to stop and interact with each other, right? So there's no real sort of long distance communication, aside from just a general alarm pheromone type behavior that ants have. Instead, when they're in the nest environment, typically when they reach another ant, they stop and they wiggle their antennae. And that intonation process takes a certain amount of time. That intonation process causes two ants to stop in a tunnel. And so irrespective of how big a tunnel might be, you still see the formation of these jams just because there's a group of ants who are stopping and interacting with each other and another group of ants coming in who will also stop and interact with that group. So interestingly, you know, even if you have quite large tunnels, because of the process of kind of trail formation and the process of stopping and interacting, you still have these fundamental limits on the rate of ant flow within tunnels. And because my talk blew up to about 80 slides, I'm not going to go into that, but I'm happy to talk with anybody about this sort of traffic flow dynamics that we see with the ants later on as well. But what I wanted to talk about more in depth now is the aspects of locomotion within these tunnels. I'm particularly interested in locomotion in which animals are sort of challenged from a mechanical and sensory perspective of moving quickly through environments that have some sort of complexity to them, substrate complexity for insect flight, maybe some three-dimensional complexity. And I think tunnels present one of the most challenging environments that an animal can move through. They're moving through an environment that has incredibly rough ground. They're moving through it, potentially climbing vertically. They're moving in pure darkness, so with no sort of visual feedback in these environments. So what do they do? And so we wanted to study this. We studied it by performing high-speed video experiments, looking at animals, individuals moving vertically within tunnels. And this is a sort of prototypical or standard descending climb of an ant. And what you'll see is that I'm just plotting out the Y position here and the tilt of the body. But what you kind of notice is that when I slow this down, you can see two pretty clear slip and fall events. And that may seem sort of innocent, but I found it really interesting that one, just a casual observation of these ants moving up and down tunnels, casualism, you come into the lab and you take a look. What you'll see is that they slip and fall all the time. So movement up and down these tunnels is not necessarily this process of precise placement of limbs so that you can generate all the required forces so that you don't slip and fall. It potentially is a more haphazard. I'm putting my limbs where I think they need to be. If they don't hit the right spot, I might slip and fall. But as we saw here in the case of this particular tunnel, when they slipped and fell, nothing detrimental happened. But that seems to be related to the size of the tunnel that the ants are moving through. So we wanted to test this and ask this question of how does the size of a tunnel that these ants create affect their locomotion capabilities just from an individual perspective? And so we did an experiment, hold on, this thing keeps jumping ahead, where we basically set up a foraging arena and we set up a nest. The other thing I love about working with ants is that they're a fantastic species to work with in the lab. You can basically get them to do whatever you want. You pull a colony with a queen, you put them in a nest environment, you put food and sort of a foraging arena with water somewhere else. You put between those two sources whatever your experimental protocol is going to be. And you get ants freely moving between those two things. And you know with some care and experimental setup, you can collect a lot of data and you can do very careful manipulations of them. So this manipulation was looking at the sort of selectivity of moving through different size tunnels, the ability to move at different speeds within different size tunnels and the propensity to not slip and fall or to arrest one's sort of haphazard locomotion within different tunnel sizes. And to make a long story short, one of the first things that we saw that I thought was really interesting was that we still see lots of slips and falls. So these ants are now moving through glass tunnels, glass tubes, smooth surfaces. Ants have these viscous adhesive pads on their feet so they can climb vertically and inverted on these smooth surfaces. Yet they still slip and fall. And it happens naturally. We're not inducing anything here. It's just a sort of again casual observation of ants slipping and falling. These are high-speed videos slowed down. And one of the cool things I think that we found from this was that I'm just highlighting two of these sort of same behaviors here. When they're slipping and falling head first, you note that the antennae are the two appendages that are hitting the wall first and seem to be the things that are pitching them into the opposing wall and sort of enabling this recovery process. And I think that this along with the observation of these ants using antennae to construct grain assemblies for load carriage seem to suggest to me that antennae are far more than just these sort of sensitive sensory appendages. They seem to have more mechanical tasks to be used in a sense of supporting your whole body weight to catch yourself as you're falling, to manipulate objects, things like that. And that's something that hadn't been reported in the literature, which I think is really, really interesting. Beyond that, we wanted to now look at the sensitivity of how a tunnel size affects an individual's ability to recover from a slip and fall. Okay, so again, you're moving through tunnels, you're moving in darkness at high speed, potentially at high speed, slipping and falling. So how does the size of a tunnel affect that? We induce slips and falls through a vertical perturbation. So putting these tunnels now on a short displacement stage that just moved at a relatively high acceleration. It says no ants were harmed in this experiment. I don't know if that's entirely true. Ants might have been harmed in this experiment. And the experiment looked like this, right? So this is, you know, in real time, basically. So we're just knocking them from the wall, totally unphysical. I'm not simulating earthquakes. I'm, you know, basically just saying in the most extreme case what happens when you're kicked from the surface of the tunnel. And the result that I'm going to show you is one that I think everybody could have guessed ahead of time, right? So why did we do this experiment? What you see is that if I normalize the tunnel diameter by the ants' body length, what you see is that for small enough tunnel diameters here, we see that you basically stop your fall. You don't fall all the way to the bottom of the tunnel at 100% success rate. As that tunnel diameter becomes larger and maybe becomes approximately two times your body length, you basically always fall to the bottom of the surface. Okay? So at some level, not that surprising, right? You basically are able to arrest a fall if you catch yourself in a tunnel or if you fall in a tunnel that enables some element of your body to span the whole tunnel diameter. Right? And I think that's something that, you know, we all could have potentially guessed. We do the experiment. We see that that tends to be the case. But we were interested in understanding what the, you know, how this might relate to the size of tunnels that these ants preferentially create in their natural nests. And so to do this, we performed X-ray computed tomography experiments on actual constructed tunnels within, you know, freshly wetted, nice monodispersed grains of glass. And we allowed the ants to construct these tunnels for, I believe in this case, it was 48 hours. And then we put them in a CT scanner and we measure the tunnel diameter. You see something like this. So, you know, they create these vertical tunnels. Vertical tunnels then create these lateral offshoots. The tunnel diameter does increase over time. And the tunnel diameter in these cases when they're initially creating tunnels, we think is the, you know, sort of this trade-off between being able to squeeze yourself into the tunnel. But also to excavate at speed so without the expense of excavating wider tunnels. And what we found is that the size of the tunnel diameter was relatively consistent across numerous different replicates of this experiment. So you basically repeat the experiment with many, many different colonies. And what we find also is that the tunnel diameter that they tend to prefer to create is also at this sort of approximate body length diameter. And I'm, you know, I think maybe sort of walking way out on a plank right now by showing these two plots next to each other. I'm not necessarily arguing that ants are constructing tunnels that enable them to arrest slips and falls. But what I do think is interesting is that they're constructing tunnels that I think are conducive to their own locomotion. Okay, so, you know, because we're all friends, I'll sort of give my analogy that, you know, when we construct steps, we construct steps that are related to our typical body size and our typical step length, right? It's uncomfortable for me to walk up a, you know, a pavilion where you would be sitting on it and the step height would be two and a half times the typical step, right? So we construct locomotor environments that are conducive or that are related to our body morphology. And I think the ants are doing the exact same thing, not necessarily because of slips and falls, but also maybe because of just the preferential arrangement of limbs as they're climbing up these surfaces or down these surfaces. So what we're trying to draw is this connection between the sort of ability to shape an environment towards your own morphology, which may enable different sort of locomotion functionalities, which I think is particularly interesting. And I'm interested in sort of continuing on in the work that I do. One of the other things we noted was that the posture during climbing changed. So, you know, ants are moving, in the experiment, moving through tunnels that are up to three times their body length in diameter, and all the way as small as about half their body length in diameter. And what we see is that the typical posture that they prefer when they're climbing in tunnels is this sprawled mid-limb posture, so keeping the limbs, the mid-limbs as sprawled out as possible. And that mid-limb sprawl posture or length is about a body length, just to give you some perspective. So, you know, they're basically their body length just in horizontal distance. And so if you look at the mid-limb distance now as a function of tunnel diameter, what you see is that as the animals become more and more cramped, as the tunnel diameter becomes smaller with respect to their body size, what you see is that basically they become more cramped. And we just quantified this sort of posture of their limbs in this cramped climbing behavior just by characterizing whether the mid-limbs were arranged so that they were pointed towards the frontal direction or if they were pointed back. And so we call this basically a sprawled or confined posture. The confined posture being with the mid-limb pointed backwards, sprawled being sort of able to sprawl their limbs out. What you see is that in these smaller tunnels, basically, they always preferentially are using this posture where the mid-limbs are pointed in the backwards direction. And this gets at another aspect of antelope motion that are in particular, I'll say, insects that I find fascinating, which is that they have these incredibly complex limbs, these limbs that have spiny hairs distributed along the length of the limb. They've got these tarsal claws, a adhesive pad at the end, spines throughout the limb. And so when they're climbing in this more sprawled posture condition, these claws and this adhesive pad can engage the surface, they can generate all the traction that they need. When they're climbing in this more confined posture, they're able to use the spines, these spines that are distally pointed away from their body to engage the surface and climb and generate traction forces via those spiny hairs. And we've seen this before in a study of cockroaches moving across wire meshes where the similar spiny hairs that are distributed along the lengths of these cockroach legs are able to engage these mesh surfaces, enable them to move at basically the same speed, pretty close to the same speed as they do on flat ground. And so I think that one takeaway for me is this idea that the legs and the appendages of insects are highly multifunctional. You have the ability of antennae to act as mechanical appendages. You have the ability of these distributed spines and these sort of multifunctional tarsy with claws and adhesive pads to engage all sorts of different surfaces to contend with whatever sort of complex environment they might be dealing with. And so, you know, this brief look at subtraining locomotion, you know, I really focused on the tasks of individuals moving through these subtraining environments and how this relationship between the size of the environment that they create is related to body morphology, potentially for locomotor purposes. But also I think that there's sort of interesting examples again, like I said, of just the different mechanical uses of these appendages. Any questions on any of this work so far? David, yeah. Is there some kind of negative feedback when the tunnels get too wide? You said that maybe ants are not trying to build that size, but maybe just naturally ride like a kind of negative feedback curve when they get too wide. I mean, I think that that's a good point, right? So that's where I said I sort of was walking out a very long plank and somebody could have sawed it off on me like David right here. Because I think that trying to say that this size of the vertical tunnel that they're making is explicitly for individual locomotion. I mean, that can't be true because the plaster cast of the first nest that I showed you, I mean, that's a nest that was probably three or four years old. The vertical tunnels in the center of that are about three to four times larger than the tunnel, preferential tunnel size that I showed you, right? So the creation of these sort of initial tunnels for a nest are, I think the trade off there is, try and get as deep as you can as quickly as you can. And it's not necessarily do that with some excess safety factors so I don't slip and fall. I mean, the other thing we should note about insects is that, I mean, they can slip and fall all they want. I think it's going to happen, right? So you can throw them off a building and they're going to be fine. So what is slipping and falling really mean? And that's where I think that, again, you know, a caveat to this notion of movement through these tunnels of their exact body length in diameter isn't necessarily to inhibit slips and falls. I think it may also be just to give them the best possible ability to engage surfaces with all of their limbs across the different sides of the tunnel. So the main tunnel is much wider? Yes, yes. Yes, and I mean, I think the other thing that I note is that in the forging tunnels that are horizontal, right, slipping and falling isn't an issue. And those tunnels, certainly the tunnel diameter gets larger as you get closer to the central nest, kind of much like a river system. You know, you have basically the most distally arranged tunnels are the smallest and they get bigger and bigger and bigger in diameter as you get towards the center of the nest. Yeah, go ahead. What happens in terms of information transfer when two plants meet and you have any suggestions to what kind of... I mean, I should ask you that, I think. No, I don't. I think that, you know, what they're doing as far as I understand is basically there's waxy cuticle and they're, you know, sensing whether it's friend or foe. You know, whether you're one of my kin or if you're a neighboring fire ant colony. Again, these, you know, they live in high... The neighboring fire ant colonies don't seem to negatively interact with the central nest. Well, that's true. I mean, so, you know, you've got monogen and polygyne, right? So multi-queen and single-queen fire ant colonies. And so these sort of multi-queen fire ant colonies, I don't quite know the distribution of them. But yes, no. So, I also don't know. So, yeah, go ahead. Why are they walking in a regime where they fall so much? Like, I mean, is it possible? What's causing them to fall? What if you have six legs and all these attachments? They should have enough fire power not to fall. I mean, I think that's a good question. I think it could be a matter of who cares about falling, right? So why take the time and effort to place your feet so that you can exactly, you know, balance your initial acceleration by all the forces you can generate with the ground? Maybe slipping and falling is just a, you know, it's a feature of moving in these sub-train environments where you have surfaces all around you that you can catch yourself and, you know, arrest your fall. So I don't know. I mean, I think that in these experiments, you know, some of them are triggering an alarm. Some of them are just doing their normal behavior, right? So the slips and falls in the glass tunnels that I show are just, they slip and fall. It could be a smooth substrate. In these tunnels here, I think in these particular experiments I was inducing some higher speed movement. But even then, you know, the speeds that I saw were not that different from just their typical up-down speeds. So I think that as we start to look closer, and this will feed well into the next part of the talk, I think as we start to look closer, limb placement error and slips and falls, I think are a very common feature of insect locomotion in realistic environments, right? Not a smooth glass microscope slide. And I think that it's just not detrimental. So maybe there's no need to build in all that extra complexity of motor control. And I... Within a tunnel or just in general? And you're talking about speed or... You know, I haven't... I've worked basically two species. So from my own experience, you know, the typical speeds are around 10 to 20 body lengths per second. So across, you know, the size variation of speed that I show in my initial slide, I don't know quite how that varies. Sure, you're thinking... So you're talking about fruit number and, you know, relationship of kind of inverted pendulum models of leg and locomotion. I, you know, I haven't done much analysis in that sense. I also think... I think we're going to hear... Maybe we'll hear some of that on Sunday... I'm sorry, on Friday. But I think that applying those kind of models to small-scale hexapods is also, I think... Well, I think it's an interesting area of research right now. I think that it's not necessarily clear whether inverted pendulum models or spring-loaded inverted pendulum models are appropriate for these small-scale animals as well. And I'm, I think, happy to have many discussions about that while we're here. Let me move on. So... Oh, yeah, go ahead. Sure. Do they have any... Yeah, yeah. So I'm sorry I didn't maybe give a more primer on ant locomotion. Ants typically walk with an alternating tripod gate. So to fore and hind limb, mid-limb on the opposing side, and they just basically alternate between those. They're step frequencies. Well, we'll see a sort of more clear example of just horizontal locomotion in the next part of the talk. But alternating tripod gate. And that's a typical fast, fastish gate for most hexapods as well. So, you know, it's hard to analyze whether it's explicitly alternating tripod in this tunnel right here in the smooth glass slides. Yes. So we've done that. And they're sort of doing this alternating tripod gate. Their antennae are, I mean, don't appear to be really correlated with their sort of stepping patterns, although that's something that I'm really interested in. The ants are using basically their antennae as the only real sensor of what's coming up ahead of them, particularly underground environments where they don't have a vision, right? And so whether their antennae are used to inform the placement of their footsteps, I think that's something that is really interesting that I want to keep working with. We don't know. But I'll show you some... Oh, no, go ahead. Totally. So in horizontal tunnels, in natural nests, there's a large variation of tunnel size as well. And again, it's sort of smallest at the furthest away from the center of the nest, larger as you get closer to the nest. It's larger and more oval in the horizontal tunnels, right? So instead of just being circular, you get now these more oval-shaped tunnels, as you might expect, just to maximize the kind of floor space area. And we did do a number of traffic studies. I have, you know, I can talk about that. I'm happy to after. So is there any correlation between the size of the tunnels and the traffic that ants produce new kind of channels? You know, I was expecting to see some... I was expecting to see, you're right, hypothesized that we would see some preference for the larger tunnels. We didn't really see any preference. We didn't see any preference in terms of the number of ants that were going up and down each tunnel. And again, I think that's related to some of this traffic work where you find that even in the largest tunnels, you still experience what we call, you know, traffic jams because they're... It's this necessary process of stopping and interacting with each other, right? Even in, you know, if you want to think about the ground as a sort of infinite diameter tunnel, they're still basically walking along these paths that are defined by pheromone trails. And so they're still always running into each other and always clogging up. And so the tunnel diameter does come into play eventually. But I think that before that, you've got all this sort of behavior of just stopping and interacting that's causing problems for them. But it's also necessary. So I think it's sort of this trade-off. I'm happy to take all these, you know, or continue these discussions after as well. I want to talk a little bit now about sort of moving above ground and looking at locomotion on rough grounds. Okay, so in particular, this is work that my postdoc has been doing, looking at how does substrate roughness affect speed? So, you know, the answer found in almost every continent. They live in all sorts of different substrate types. They live in trees. They live on the ground, underground. And we're interested in understanding basically how the complexity of these substrate environments affect locomotion. And a number of people have done work in this area. So I'll note that, you know, I'm not the first person to think about this. And folks have looked at ants forming trails in the presence of a smooth ground and are sort of more challenging ground. And they tend to form these trails that minimize the total time it takes to get from point A to point B when, you know, they're sort of in the presence of a substrate that's going to slow them down. People have looked at the effective tree bark roughness on arborial ants. Jerome has looked at and been looking at for a while locomotion on granular substrates. We're going to hear from him a little bit later today, not necessarily on this topic, I think. And so, you know, people have been thinking about this for a while, how substrate roughness affects legate locomotion. And we wanted to, you know, sort of attack this from a large scale perspective, just how are these ants affected in terms of sort of macroscopic speed and things like that. But we also then wanted to understand more specifically what's happening at the foot placement air perspective. You know, how are individual feet affected by roughness substrates, different roughness substrates. So we did an experiment where we took an Argentine ants into the lab. So we don't have, we do have fire ants in San Diego. They're a different species, not Solonopsis and Victor. But more prevalently, we have Argentine ants who are basically, they live in these super colonies. They're incredibly easy to find. You just step outside. You can dig up a colony, bring them into the lab, set up another one of these experiments where you have a colony separated by some substrate that you might want to look at their locomotion abilities on, and then a foraging arena where you place food and water. Set up a nice trail going back and forth. Use cameras to image them as they're walking across these substrates. We used basically three different types of substrates and then one flat substrate. We 3D printed checkerboards. We chose this as a substrate that had a single length scale. So it was one millimeter in height. But then the length scale in the horizontal direction was one by one, three by three, or five by five. We then monitored them moving across these substrates. We did this for, I believe, 10 colonies. Again, I don't quite remember the exact number. We did this for about 11,000 ants. And we recorded about 200 frames per second. So we can get out very nice information about the whole body movement as well as where their individual limbs are being placed across these substrates. The first thing that you see is that ants establish these pheromone trails. So just on all the different substrates, you see pretty nice distributed paths that are localized around a single pheromone trail. No real clear pattern or difference of pattern of these pheromone trails across the different substrates. I think kind of interestingly, and again because we're all friends here, you tend to see that there does seem to be some sort of cross cutting of the five millimeter length scale. I can say that the five millimeter length scale is about the size of these voids here. So they do maybe do some edge following on these larger substrates. We measure velocity across these substrates of the individuals. We then take as a measure of this sort of velocity of a trail. We take the median value of that. We aggregate that. And we measure basically the median speed of all individuals across these different substrates. And so what you see is that basically these ants are walking at about 15 millimeters per second on flat ground with a relatively wide distribution in terms of their top speed. Now again, this is sort of the speed. An individual point in this distribution is an aggregate measure of a trackway. So an individual walking across the full substrate. We turn that into one measurement of the median speed of that path. We then aggregate that into these distributions. So what you see is that on the one millimeter substrate, the speed is, it's not quite half, but it's at about I would say eight millimeters per second. And then you see a subtle rise in speed as the substrate gets coarser and coarser. So again, flat, one, three, and five millimeter in horizontal spatial scale. So first observation, which mimics what other people have observed, roughness does affect individual walking speeds. We see that there's a specific length scale dependent effect on this walking speed. And so the next question that you can ask is, is this a speed limit in effect or is it a result of behavior? Are the ants just slowing down because the substrate is foreign to them? Because for whatever reason. We attempted to test this by basically inducing a more alarmed state. And so you can do this by basically pumping in a little bit of cinnamon infused air. So you know, you go out during Christmas time and you find like some cinnamon potpourri. You put it in a solenoid and then you just allow it to be pulsed into the tunnel. You do it once every hour. You can then get a before and after perturbation speed. And we attempted to measure the top speed. So before we're looking at kind of the median of that distribution, now I'm going to look at the cutoff for the upper 95% of the walking speeds. And I'm going to look at a before perturbation, which is the circle, and an after perturbation, which is this line here. So what you see is that the sort of peak speed that they're walking at basically is relatively slower on flat than when they're stimulated. So there's this increase in speed. When you apply the cinnamon perturbation, they increase their speed on flat ground. Same thing happens on five to a lesser extent on three and to the least extent on one. So it looks like the top speeds on these one millimeter substrates are pretty much not behaviorally modulated. They're controlled by just the ability of the ants to walk across these one millimeter substrates. We also wanted to ask the question, is this ecologically relevant whatsoever? So we're collaborating with an ecologist who's very interested in Argentine ant dispersion. Where do we find them? How do we stop them from getting there? And so we did field tests at the UC San Diego Field Station where we put out the same substrate. We allowed the animals to establish pheromone trails to a food source giving the choice of rough substrate or flat substrate, but also giving them these three different options or three different trails. So one, three, and five versus flat. So we could test the effective where we find a pheromone trail being established or a forging trail being established versus flat ground. And so in the case of the one millimeter substrate, what we find is that there is a highly significant preference for walking on flat ground. So about 75% of the time ants establish a trail on the flat ground substrate. As you look at the three millimeter, you see that there's also a significant effect, but it's a slightly wider variation. On the five millimeter, no significant effect. So we found pheromone trails equally on flat substrate as on the five millimeter substrate. And this seems to match again our observation that the most effective substrate at inhibiting walking is this one millimeter substrate, which again presents them with the most perturbations per unit distance anyway. So maybe that's exactly what we'd expect. We wanted to take this a little bit further and ask underlying these speed changes what's going on at the limb level. So are they changing their gait? Are they using their limbs in different ways? And it just so happened about a year and a half ago, this open source deep learning package was put up on bio-archive to enable sort of automated tracking of animal sort of skeletons, jointed models of animals in an automated sense using this deep learning toolkit. There's now a number of these deep learning toolkits for doing this automated tracking. Leap in particular was nice because it's specifically set up for observing animals in this dorsal field of view so you can use some of the symmetry properties of the animal from that perspective. So we built up a deep learning model. We applied it to locomotion on these flat substrates. See, it works incredibly well. This leap toolkit we just trained on the antennae and the distal ends of the feet. So we weren't tracking every joint. We were just focusing on where the feet are placed and what the velocities of the feet are. Leap gives you an estimate of the confidence of the measurement so you can do some sort of data cleaning and things like that. As well as working well on the flat substrate, we were a little scared of running it on the rough substrate, but it worked relatively well as well. And here you can see our first glimpse of what's potentially different about locomotion on these rough substrates. You can see it's slower, it looks again less rhythmic for lack of a better word. And there seems to be maybe more error in foot placement. We looked at the first measurement that people typically do with lego locomotion, so kind of a fundamental kinematic relationship. The relationship of speed versus stride frequency. I'm sorry if this is the font is small, but this is the average speed of an individual stride. So from touchdown of one limb to the sequential touchdown of that same limb and what the frequency that they're walking at is. This is flat one, three, and five. And again, the sort of fundamental relationship here for bipedal, even alternating tripod hexapods, is that you see a nice linear relationship when you have a constant stride length. Modulation of speed through changing your stride frequency, either speeding up or slowing down your stepping pattern, but keeping the same touchdown rotation of your limbs will give you this kind of behavior. And we see it's basically the same for all of these different substrates. And I say basically the same by drawing these lines. We actually built a model to predict the walking speed based on this constant length model. And we found a good agreement between a majority of the data set for the rough ground. So for the rough ground here. But this model allowed us to classify whether there was sort of an inlier or outlier, whether this constant length model described the velocity that we observed on these rough substrates well or not. And what we found was that when you build this model and you look at the number of points that are not well predicted, they lie outside of say five standard deviations of this model prediction. What you find is that this one millimeter substrate had the most number of outliers. And I'm going to call those maybe disrupted strides. So strides that don't fall along this sort of simple constant length kinematic walking behavior pattern. So again, kind of echoing what we've seen in the velocity results, one millimeter seems to be the worst in that many of those strides, about 20% of them are a little bit over. What was that? Oh, I'm sorry. Close to 15% of them lie off of this constant stride length pattern. You can see this as well as if you measure the stride to stride distance that an individual stride takes. So kind of like what's called a return map, looking at a periodic behavior and measuring the relationship between the n and the n plus one value of that measure. So the measure we chose was the length distance of a stride. And what you see again is that on flat ground, if points lie along the 45 degree, you have a nice rhythmic behavior. It repeats itself, repeats itself, repeats itself. But we see this cluster of points. I could turn the lights off to maybe show you this better. But this cluster of points that fall outside of this, that fall short. So there are these inhibited steps. They tend to be shorter than the preferred stride length. And again, you measure the propensity of those points to lie outside of this sort of repeated walking pattern, this sort of undisrupted walking pattern. And again, we see that this one millimeter seems to inhibit them the most. We can do other things. We can measure the propensity to, or the sort of foot touchdown error. So we measure the foot touchdown error of these sort of inlier points, points that are undisrupted. And you see basically that there is variation in the touchdown locations. The gray is the convex hull of all touchdowns. The ellipse is basically the characterizing one standard deviation of the touchdown locations. But if you look at the outlier points, you see that there's quite a bit more foot touchdown error. And what's kind of hard to maybe see from this blast of figures, I'm sorry, is that all of the foot touchdown errors are shorter than the typical touchdown error. And that can happen from a number of reasons that I'll detail in a second. But the point is that the foot touchdown error is larger on these outlier strides. And we can show that even more clearly by basically just building a simple predictive model that says what is the relationship between the amount of touchdown error from the preferred touchdown location and the likelihood that a point doesn't lie along this nice kinematic walking pattern line, which are all of these gray points. What you see is that logistic regressions explain this relationship relatively nicely. And there are confidence intervals within here. So they explain these relationships quite nicely. And it makes sense from what we might expect perspective that the larger the foot error, the larger likelihood that that's going to be associated with some sort of temporal disruption of touchdowns. So basically what we're finding is that these 1 millimeter substrates seem to induce the most foot error in terms of placement of touchdowns. And that foot error is associated with some kind of disruption of the temporal walking pattern, which there's a cause and effect thing that we don't necessarily can't necessarily explain, tends to cause them to walk at slower speeds. The slower speeds, though, that you see, the dominant behavior is they're still walking like they're walking on flat ground. They're just walking at slower frequencies. So all this could have been explained, maybe, in just saying 1 millimeter substrates, they walk with slower stride frequencies. The question is, is that happening? Because as they go to higher frequencies, they see or experience more foot error, which then causes these temporal disruptions, which causes them to slow back down. Or is it just already a sort of modulation of their stride frequency because of the substrate itself? So let me ask this or address this question now. What is an outlier? So what is one of these disrupted strides look like? I'm going to turn the lights off just to see. Nope, all the ones that I didn't want to. Oh, that's fine. OK, maybe we can see better. So what I'm showing now are just a selection of these quote unquote disrupted strides, strides that had a large touchdown error in terms of location and that were associated, like I said, with these temporal disruptions. And what I want you to notice, because it is subtle, is that I'm highlighting a limb here. Red is when it's touched down. When it goes back to white, that's the end of that quote unquote stride. What you'll see in all of these cases is that these are instances where a limb is being put forward and hitting an obstacle ahead of when it would have preferred to a touchdown and its preferred touchdown location. So we're seeing instances where these limbs are basically moving forward, hitting an obstacle, pausing for a second, slowing down. Our algorithm of measuring touchdown location is based off foot velocity. When the foot velocity goes to zero, you measure that as a touchdown. And we see these instances, again, where a foot is moving forward, hits an obstacle, and then gets to the preferred location. So moving forward, starting now, hits an obstacle, and then is potentially repositioned. And whether this is stopping because behaviorally, they stopped their limb and then they repositioned, or stopping because mechanically, they can't move their limb past that obstacle based off the force that they're generating from the muscle. We don't know. But we're seeing is that one millimeter substrates are inducing the most of these types of behaviors, where limbs are haphazardly, I'll say, hitting obstacles during the swing phase and either being repositioned purposefully, or just getting to the end location based off of just the compliance of the limb and the springiness and just moving your limb through that collision. So we see these kind of inadvertent limb collisions, which we think are the root cause of most of this behavior of foot placement error, which we also think is the root cause of these speed reductions. And so the conclusion of this is that we think ants are using these similar limb kinematics on different roughness substrates, experiencing larger and larger foot air, touchdown air, on these different substrates. And that's the sort of what's explaining the speed change on these one millimeter substrates is this sort of foot touchdown error. Okay, I'm going to blast through this very, very quickly. What I really just want to show is that basically there's lots of inspiration now that insects have been providing for roboticists, particularly in the areas of mechanisms, control, navigation, and collective behavior. And in my lab, we're actively working on building mechanisms and building new types of robots, specifically new types of fabrication methods for robots that are inspired by the insect exoskeleton. And I'm just fascinated by insect behavior, but also by just the mechanical properties of the way that they're built. So I think exoskeletons are incredibly interesting. They're a continuum of stiff and flexible materials integrated together, interesting sort of under-actuation properties, passive mechanics that enable locomotion, and fascinating sort of nonlinear mechanisms. There's kind of two ways in which we do this in my lab. One is there's a very expensive, fancy way of making these insect scale robotics that was really pioneered at Harvard by my postdoc advisor, Rob Wood. You know, you use a laser, you cut out these precise patterns in rigid and flexible material. You arrange them in such a way so that when you stack them together, you can make mechanisms. These mechanisms can be things like transmissions that can turn one motion into another motion, and you can build robot legs and robot hinges. You can build these at very small scales and try to approach the sort of small scale of insects. And we can study locomotion properties in small scale systems. The other way that we're sort of trying to make these insect-inspired robots, which I think is also incredibly fascinating, is a much more low-cost version. So basically, we've developed, I'll say invented this method of 3D printing with a traditional 3D printer onto heated thermoplastic films. And this is something that I really just want to get through because I think it is incredibly accessible, and anybody can do it if you have one of the even cheapest of cheaper 3D printers off of Amazon. You just need a heated bed. You need to be able to buy some thermoplastic. Again, thermoplastic is incredibly easy to find. You heat it up to a certain temperature and you 3D print on it. You can make these very flexible structures, which typically 3D printing flexible materials requires either multiple materials, it requires sort of poor choice of material and very expensive printers. So we've characterized this process. We can get good adhesion to these films. We can control the stiffness of these elements, and I'm sorry I'm going so fast. I'm a little late. And we can make structures that are something like this. So basically, these very simple structures that have mechanical functionality programmed into them, so say hinge components that jam so that we can actuate two hinges with a single tendon. We can build these jamming elements that can jam in different directions. We can build these highly flexible limbs. These flexible limbs can have maybe jamming elements that jam in flexion, that jam in extension, and that also exhibits sort of torque reversal mechanism. So with a single tendon, you can actuate very complicated motions. And all of this, again, is done using very low-cost 3D printing techniques. I'm happy to talk with anybody after about that as well. So again, my many, many conclusion of this is I think that insects provide a wonderful insight into mechanical design of these kinds of interesting structures. And my overall conclusion is that I'm happy to talk with anybody for the next 72 hours about why ants are awesome and what we can learn about them. So thank you.