 My lab is interested in is how do animals make decisions? Now what do I mean by decisions? Take this example, for example, take a bat that is in its roost, a few minutes after sunset. And this bat now has to make many decisions, decisions such as where to move, how to move, when I say how to move, I mean to move alone, to move in a group, which trajectory to use. This was mentioned already in the previous talk quite a lot. Now when making these decisions, the bat has multiple types of sources of information. It can use its own sensory information. It can use information from its peers or from other bats, what we call social information. And it can also rely on a higher cognitive skills or capacities such as memory, for example. So what we try to understand is how does a bat integrate these different types of information into decisions which are usually measured by movement. So I will start by talking about sensory information. This is also where I started. I originally was a sensory ecologist. And the main reason for me to study bats is this, is echolocation, the fact that bats sense the environment through sound emission. This is very interesting from the point of view of the animal as I'll elaborate in a second, but it's also very useful for me as a scientist who wants to study animals in their natural environment. And that is because if you look, so what you see here, I guess I don't have to explain echolocation, right? Bats or biosonar, same, same. Bats emit sound, echoes are received, the brain processes. These echoes and many different types of information can be extracted. I will elaborate a little bit later. But when we look at a sequence of calls emitted by a bat, so this is what we call a spectrogram. You have frequency over time. Each of these is an echolocation call. Just by looking at this sequence of calls, I can say quite a bit about the behavior of the animal. You can notice, for example, how the signals change a long time, their intervals between them become shorter, their bandwidth increases, the duration of each signal decreases in this case. And this is because the bat in this point is attacking prey. So I can say this just by recording the bat. If I have an array of microphones, such as in an acoustic room, for example, I can do much more than just recording the signals of the bat. I can record, for example, it's what we call beam steering, or where it is allocating sensory attention. So you can see a bat here trained to land on a target in an acoustic room. And these lines, so you can see it's trajectory in blue, but these black lines that appear whenever the bat emits sounds, they pick the center of the beam of the bat. So you should imagine this acoustic beam, like a beam of a flashlight, for example. But of course, we're talking about sound waves. In this case, and it has some kind of directionality. And apparently, the bat can also control where it directs its energy. Now, this is in general a very powerful tool for me to assess what the bat is attending sensorically. In this case, we also show that the bat, instead of pointing its beam straight. So you could see it here, maybe better. Instead of pointing its beam straight to the target, what the bat was doing was pointing its beam to the right and to the left. And what we showed is that instead of pointing, so we have here the right beam and the left beam. And instead of pointing the center of the beam towards the target, it points the maximum derivative of the beam towards the target. And by this, it actually reaches maximum sensitivity to changes in azimuth of the target. You have a question? Sorry, just to add. Perfect. Exactly. So you could think of this. So these are all microphones. You could think of each of them as a pixel, if this was an image. And we reconstruct, in this case, it's 1D. So we reconstruct the cross section through the beam of the bat, which is what you see over here. We now have a 2D array covering the entire all of the room. And what I show you here is you see this point within the beam is pointed towards the target. And this is the maximum derivative of the beam, so where energy changes most rapidly over angle. And by using these, so it can do this with one beam, but by using two beams, you double the derivative, so to say. So this is one strategy used by bats. Another strategy used by bats in order to improve their sensory acquisition is the following. In this case, we have the same array that you've seen in the previous slide, but here we put it in the field. So you can imagine a small pond of water. So this is the Israeli desert. The pond is two meters by two meters, something like that. And the reason why we have the array, these circles here, the red circles depict our array. The reason why we have the array here is because we anticipate the bats. So they come to drink. And we can therefore record and reconstruct their beams. Remember this point, because remember how limited we are. Later on, I will show you how we overcome this limitation. We can only record the bats with this system when they come near our array. But if a bat indeed approaches our array, we can reconstruct its flight trajectory. You can see here in blue, a reconstruction of movement based on sound only. So every time the bat emits sound, the same signal will arrive at the different microphones with different delays. And this allows us to reconstruct the position of the bat. And in this case, I also show you the beam of the bat. So not only the steering, as I showed you before, but an actual reconstruction of the beam. And what you can notice is that the bat changes the width of the beam. So this would be equivalent to changing the field of view. For us humans, for example, you can see that when it goes down, it narrows its field of view. And then when it emerges from, so this is a small pond in their banks. So it's going down between the banks and then flying out. And it increases or widens its field of view. And we can try to interpret this. Perhaps it's trying to avoid echoes from the side. We can, there are many guesses for why this sensory behavior is beneficial. But how does it do this? It does this just simply by changing the aperture, changing its mouthgate. That's all it has to do. And we actually prove this by using imaging, by photographing the bat while they do this. You can see that a wide beam corresponds to a very narrow gap. And a narrow beam corresponds to a very wide gap. So since many people here do robotics, I want to show you our take on robotics. We're much less interested, or we know much less about the motor side, which is what I think most people here are doing. We're much more interested in the sensory side. The students, by the way, who are in charge of each project are always on the slide. So here is our autonomous acoustic vehicle. It is navigating through the Botanical Gardens in Tel Aviv University. It has a camera just for us to see where it is. There's no use of the camera. It's fully autonomous in the sense that everything is being computed on the robot itself in real time. What it does, it emits sounds. You can see two ears. There are two microphones. So there's a left ear and a right ear. And whenever it detects an echo, it localizes it based on these two incoming sensory inputs. It can then reconstruct a map. Again, this is based on the acoustics only of the Botanical Gardens. So now you can see it planning its path through the map that has been reconstructed. And it can even perform what I call simple decision making. So for example, if it ends up in an area like this, where it is blocked from one side and blocked from in the front and onto the right. And on the left, there's some plants. So what the robot is now doing is to acquire echoes from all directions, run these echoes through a neural network, trying to find which of these paths is free. Or in other words, where is there a plant that it can go through? And as you see, we're not so good in the locomotion. We're better in the sensing. But in a second, it will manage to go through this plant. And everything is fully autonomous. How can we classify objects based on sound? Maybe some of you are wondering after I showed you these images. So here's some intuition to how we can do this. What I'm showing you now are spectrograms, not of a bat, but rather of echoes. There are five species of plants that are presented. You can see one example of each, once again, frequency over time. And we ran this data through a support vector machine. This was already many years ago before the neural networks made their renaissance. And we were able to classify these different types of plants based on their echoes. And just to give you very simple intuition, I want to play back two sounds to you, the sound of an apple tree. So this is an apple tree. Of course, I have to down sample everything for you to hear. It's all ultrasonic. And this is a corn field. And I'm quite sure that you are now, for some reason, it was a bit short. Let me try to play it for like this. Seems to be trimmed a little bit, but this was better. And I'm quite sure that if I run a short test right now, I'll play one of them and you'll tell me what it is. Let's see. So I'll play this one. What is this? Corn field. So everybody is now trained to detect corn fields according to the sound. How can you do it? It's easy. I gave you an easy example, but we could do it with all of them. You can clearly see so time when using sonar equals distance. And you can see that we can look with our sonar system into the corn field. You can see the first, second, third, and fourth rows of the corn field. So and this is what you could hear. You could hear this segmented sound. Something like that. But we could also, and I can give you some more intuition to how you can classify some of the other echoes, and we could do it for all of them. Before I move to talk, yeah. You're singing one way, right? Yeah, in this case, it's a chip. It's an FM chip mimicking a bad signal. It's a very short pulse between 100 kilohertz and 20 kilohertz, modulated linearly in this case. OK, so you're on. Yeah, yeah, yeah. That's, in terms of what are we doing new with robotics, except for all of these machine learning algorithms, I think most people using, not so many people use ultrasonic robotics in air, but the ones who do use them for, for example, range finders, and they're always using these very narrow banded pulses, while we exploit the entire bandwidth just like bats do. And you could hear it when you sound, when you heard that apple tree, you could hear this, because it's, yeah, it's chirping from high to low. Yeah. So you said the bats. So it depends on the species. And some bats will actually even use constant frequency signals, and they will detect Doppler ships created by the flutter of the wings of a moth. I will not elaborate on that unless you're interested. But most bats use FM chirps. The range can be between bats that use a bandwidth of 10 kilohertz between 30 and 20, and they expertise in, for example, hunting larger insects in open space. And some bats will use bandwidth of more than 100 kilohertz, and they will sweep from 200 kilohertz to 80 or 60. There's a huge variability. It depends mostly on the environment in which they hunt. And thanks to radar literature, we can also explain quite accurately, I would say, what's the purpose of each signal. So I'm about to switch to talking about movement. I just wanted to remind you that bats are not only echolocators. They all use vision. We actually had a recent study showing that they can translate. Translate maybe is a little bit too strong, but they can create a visual representation based on acoustic information. So you can train them using echoes, and then test them using vision. And they are able to perform a classification test. But in this case, I want to show you another sensory system used by a tactile system. But I'm quite sure something you haven't heard of, and also very preliminary. I threw in these images yesterday. So some bats have a tail that protrudes out of their membrane. In most cases, the tail is part of the membrane. This is probably what the ancestral bat looked like. We call this a greater mouse-tailed bat. And what this bat will do when crawling up a crevice is to move its tail just like a blind person uses a cane, probably in order to sense the morphology of the crevice. I'll show you a moving, not perfect. You'll see the quality is not perfect. You'll see this bat crawling up this wall that we created. And here we are tracking the tail. So you can see how the tail is moving right, left, right, left. We are now analyzing the kinematics, what is going on. I can already tell you that it's a theta kind of 8 hertz rhythm. And in parallel to doing this, we're also MRI scanning the bat's brain in order to understand the somatosensory representation of this area in the bat's somatosensory cortex. Is it large? Is it small? OK, so just I don't think so. I don't think so. It also has these, by the way, reminded me when I saw the elephant talk yesterday, it also has these small hairs protruding at the end of the tail, which might be related maybe to sensing flow when it's flying or something like that. An unknown sensory system or unexplored, it has been known for many years, sensory system that some bats use. OK, so I'm going, as I promised, I'm going to switch to talking about movement and especially about how bats use social information in order to make these decisions that I mentioned. But in order to study this, we really want to study the animals in their natural environment. Most of what I've shown you so far was in the lab. And in order to study the animals in their natural environment, we had to develop these tiny sensors that can be mounted on the animals because bats are very small. So most bat species, by the way, there are 1,300 species of bats in the world. Most of them weigh less than 50 grams. The smallest mammal on Earth is a bat weighing 1.5 grams. Some people will say it's a shrew, but it doesn't matter if it's even if it's number two, it's still respect to this bat. But what we had to develop, because there were no small enough sensors at the time, we developed our own miniature GPS loggers, which also come with a microphone. That's super important for us, right? We want to record sound on the animal. At the moment, I'm still not aware of other sensors that include the GPS and the microphone together. Today, there are quite a few small GPS devices, but I think none of them have microphones. Currently, we can also perform additional recordings using these sensors. So we can record acceleration. You can see in red here the wing bit of the bat. So this is z-acceleration. You can detect the wing bit. And you can see how echolocation in blue is synchronized with wing bit. We can do physiology. We've recorded ECG on flying bats, EG on flying bats. And also, we can measure ambient parameters, such as temperature or body temperature as well. What do we get when recording sound? I think some of this you will now already recognize since I gave you an introduction. So one of the most important things for this talk is our ability to record conspecifics. So look at this spectrogram. This is a different bat with a different signal. By the way, following your question, look at this signal. Very different from what I've showed you before. It's a multi-harmonic signal. And in red, I circled calls that are weaker than the ones emitted by the bat carrying the device. These signals are emitted by a conspecific. So the same species flying near my bat. And now I can study the interactions between them. I can also study foraging. I will not talk about this too much today. But for example, here, we can see an attack. You've seen this before. The intervals become shorter. The calls become shorter. The bandwidth, in this case, decreases. Sometimes, when I study foraging, I can also say what is being attacked. For example, here, we work on a frog eating bat in Panama. We know that the bat attacked the frog because we also recorded the frog signals just before the attack, you know, the frog. The male frog is calling for the female to find it, similar to the moths with the pheromones. In this case, we know that this was the last call ever emitted by the frog because we have the chewing sounds right after the attack. So the idea is really that we can now quantify foraging success in completely wild animals. For example, we could compare maneuverability of individuals to foraging success. But there are many ideas and we're still working on this. So what I do want to show you is a few things about how bats move in groups and why they move in groups. So some bat species exhibit this type of behavior. What you see here are the same greater mousetail bats that I showed you before. What you're listening to are social calls. This is social communication, not echolocation. This was a kestrel, by the way, that was trying to grab one of the bats. This is a very dangerous moment in the life of a bat because the bats are aggregating outside the cave and they're about to swarm out. Okay, so as I said, some bat species will show this behavior, or many, maybe I should say, but not all. And these bats will now fly together for several kilometers and what happens later, we didn't know, let's say, until we place GPS devices on these bats. So I'll say very briefly that we found that these bats move together in groups throughout the night. And I want to take you, because I don't think I have enough time, I want to take you to a different system with a similar problem. So what is the problem? Maybe I'll say one more word about this problem. These bats hunt for ephemeral food. What do I mean by this? They don't eat fruit, they don't eat insects that are easy to predict in their location, is easy to predict. They hunt queen ants performing nocturnal, noctural flights, okay? So this is something that is probably very hard to predict and we also show it in our paper. And once you find it, there is a lot of food, okay? And now by saying this, I will take you to another species that has a very similar problem. This is the Mexican fish eating bat which lives on these small islands in the center of the Sea of Cortez. And why did we think that it has a similar problem? That is because this is probably the only bat that forages on fish and crustaceans in the ocean. There are several bats that will forage near the shore or in rivers, but this is probably the only one that really goes out at sea. Sometimes dozens of kilometers out at sea, I'll show you in a second. And the problem is really, really difficult. You have to find a flock or school, right? Of fish or crustaceans in this huge ocean with very few cues. So we started tracking this bat and for the first time we could say what it is doing. So where it is moving. And indeed, as I said, we found that it can fly sometimes dozens of kilometers offshore. This is one individual, different nights. By the way, you can see here nicely, I would say carefully, a levee-like behavior. You can see how the behavior changes when there is food and then you can see longer trajectories without turning, but I'll be very careful. We didn't quantify it and it's difficult to quantify levee-like movement. And there are many interesting questions such as how do they find food and how do they navigate? So notice that it looks like the bats know exactly where they're going. They go back to this tiny island, one kilometer radius over. They fly very low, one to two meters above sea level in completely dark night sometimes with moonless night. I've been out at sea. You cannot see the island from anywhere. So it is unclear to me how can they find their way back to the island? Maybe Anna will have some ideas in her talk later on. But what we addressed in this study is how do they find food in this island, in this environment? And what we found is that they constantly move in groups. So I'll show you now a movie that shows the movement of several bats. And every time, so the island is over here and you'll see the full night of each bat, mostly they do something like this. This is still a huge area. There's no scale here. But this is something like 15 kilometers by 30 kilometers that they have to scan for fish, for food. And wherever there is a white circle, this means that there is a nearby bat, a specific nearby. How do I know this based on the microphone? I already explained before. And you'll notice when there's a red circle, it means that they found food and they're attacking food again based on the microphone. But you'll notice how they constantly move with bats around them. So they are constantly part of a group. I should say that my microphone is not as sensitive as a bat. I can pick up another bat from up to 20 meters. So that's why I only have white circles part of the time. But the bats can probably eavesdrop on other bats from up to 150 meters. So if I had the ability of the bat, you would probably see white circles all of the time. So what do they gain from moving in a group? Here is a schematic that tries to explain what is the benefit of moving in a group. So again, you have to remember that we're looking for prey that is ephemeral, hard to find, but abundance. Once you find it, there's a lot of food. It's not as if you're competing with neighbors on food or there's less competition, let's say. And what we think is going on is that the bats spread as a group and eavesdrop on other bats, right? So once another bat finds the food, everybody within eavesdrop range, which could be up to 150 meters, knows that this other bat found food, right? How does it know? I showed you before. When a bat attacks food, the sequence of calls becomes very stereotypic. The calls shorten, the interval shorten, so on and so on. So we call this the bag of chips effect. Why? Because if I turn off the lights in this hall and somebody, let's say, over here opens a bag of chips, everybody knows immediately, it's anonymous, nobody knows who opened the bag of chips, but everybody knows that there is food over here. Same with the bats, they cannot conceal this. Now, importantly, the self-detection range, when you use sonar and you're searching for an insect, for example, your own detection range is extremely limited in air, okay? Because of these high frequencies, you can only find an insect from a few meters. Some people would say 10 meters, but that's probably the largest people would estimate. So you're extremely limited, but if you spread an eavesdrop, you can increase your searching area and we actually use this, commonly use the three-radie model, all right? So I'm sure many of you are familiar with this model where you have three, each individual has three-radie. So it's an agent-based model, so if there's an agent next to you within this radius, you are repelled by it, you move to the other direction, then there's an alignment radius in which you align your movement to the movement of other individuals and then there's an attraction radius in which you are attracted to other bats or to other agents and we use this, as I said, commonly used model in order to test our hypothesis that this type of searching would be beneficial for the bats and indeed what you see here is a simulation. You can see the bats leaving their island, there is the food is sparse and ephemeral, there's only food over here. They form these groups autonomously, so the only thing they do is to behave according to this set of rules that I've explained and you see that groups form, sometimes they split, sometimes they rejoin and in a second you'll see that one of the groups finds the food and then notice what happens when this group approaches the food. So each individual here can only hear individuals within 150 meters, so this individual now found food, this guy has no sense of the food being found, it's only following the guys in front of it and still we find that using this very simple set of rules, something like 80%, you see, not everybody, 80% of the individuals in the group will converge onto the food, so and we really show that this is much more beneficial than searching individually. We're also now in the process of building these swimming robots, as you see we're avoiding flight, we're either on the ground or in the water, we don't want to deal with flight because as I said, I don't know much about motors but we're in the process of building a group of swimming bots that will behave sensorically just like I have now explained, hoping to see this same type of behavior emerging based on this simple set of rules. So in the third part of my talk, this was about how bats use information in a group, in the third part of my talk, I will try to touch on mechanisms, a little bit more deeply, and I will start by talking about how do bats navigate. So the bats that I showed you before, those Mexican fish eating bats, first of all their task is to search, so it is not very clear where they are going and we don't know much about their navigation but I will now talk about fruit bats that we turn to the same fruit trees night after night and for them it is easier at least to think that we understand where they are going to study their navigation. Now the problem with studying navigation with wild animals is the fact that you don't know their history or maybe let me say, sorry, let me say, let me rephrase, the problem with testing the idea of maps in animals is in the wild is the fact that you don't know their history. So there is this hypothesis that some animals have perhaps maps, what we call cognitive maps, they can map the environment and they have some kind of mental map in their brain and one of the hallmarks of cognitive maps is the ability to move between two points, two familiar points, using an unfamiliar, relatively straight trajectory, okay? So think of yourself, you want to move from here to, I don't know, the train station in Tuest and if you've created a map of this area, you can do this with a relatively here, it will be easy because you have the shore but a relatively straight line even if you have never traveled in this trajectory before. So what do I mean by history? Here's a bat that we tracked for almost 100 days and I need to say that bats are long-lived, they can live up to 30, even 40 years, much, much longer than any other mammal with a similar size and look at what this bat is doing for something like 20 nights, it flies to point A, something like 25 kilometers away from the roost, this is the roost and then it flies to point B for several nights and then it flies to point C for several nights and then at some point around 960, it performs this shortcut in yellow, this clear shortcut. This is something like 10 kilometers and it's dark, it's night, it's hard to believe that it can beacon, that it can see point C from point B. So it looks like a map, right? But the problem is that I don't know the history of this bat, right? I caught it at some point during its lifetime, it can be one year old, well it's not one year old, I know that but it could be five year old, it could be 30 years old and I have no idea if this is really a novel shortcut. So in order to overcome this problem, we invented the following method, we developed our own bat colony. So when I say developed our own bat colony, we simply opened the window in our bat colony and we hope that the bats will come back and indeed many of these bats come back, these are fruit bats, the colony is here in Tel Aviv University, it's the first academic colony of bats in the world I think and what you can see here are three nights in the life of one individual outside the colony, each night is depicted by a different color and very fascinating for me is that already on the second night outside, this bat goes to, slips over at friends, so you can see it finds a colony nearby our colony, the area is full of colonies, a full of fruit bat colonies, they are very successful in the city, so she slips over in this colony and then comes back to our colony. But the main thing about this method is that it now allows us, when we study pups to monitor the full history of the animal, so you will now see 50 nights continuously in the life of one individual from the day it left the colony for the first time, the nights are color coded from blue to red, you can see that there are many short exploitatory, so the colony is over here, there are many short exploitatory trajectories when she flies nearby to a familiar tree and there are also often these extreme exploratory trajectories where she probably is increasing, in this case it's not a sheet, so he is increasing its knowledge about the area, but now for example when it arrives here and then comes back, I know that it has never, so if we're looking at the yellow trajectory, I know because the red is later, I know that it has never been in this region before, the closest it has been to it is several kilometers away. So in the future, we hope to charge them through induction, so they will each land on its port and will charge itself, at the moment we don't have that and that's exactly why we needed this colony, because if I now try to work in the wild, I can place a GPS, it will work for two nights and then I can dream of catching the same bat again and again and again, in this colony we can catch them every three nights for example and change the battery. So the data is logged, it is not transmitted again due to energetic considerations and again these are exactly the two main reasons why we needed this colony, we can just remove the tag, put a new tag and the bats get used to this and not all of them stay by the way, many I would say only 30% of them stay, but that's enough for us to get a good sample size. Yeah, as I said these are fruit bats, so we can also map the fruit trees where they prefer and they return to these trees often, although not always, I will talk about this later if I still have time, so we know something about where they are heading, right? We think at least that we know and here's to the main question, so do they perform shortcuts? So in black here, I apologize for the black color, you can see a novel, a truly novel shortcut, how do I know that it's truly novel? In blue you can see what this bat did during the night before performing the shortcut and so this is a forging tree and this is the roost and then in white you can see everything that this bat, the entire history of this bat, so everything that it did before and you can see that it never performed this type of this movement, so it hardly ever has been to this area before. We have detected several, something like a hundred and here it's written 125 shortcuts here, some more examples and all of them are, as I said, novel, are they straight, are they direct? So here we quantify the straightness of the trajectories of these shortcuts using the straightest index, so it's the ratio between how much you actually travel in the straight line, so one would mean straight. Commute flights, commute means the bat uses a trajectory that has already been used, so you expect it to be very, very straight and indeed you see that it's close to one and shortcuts are also very straight, as you can see over here, so the bats are flying straight between these two familiar locations. If you look at what we call exploration, exploration means you're moving to a point where you have not been before and you can see that the straightness is much, much lower and it's the same for some model in which we try to mimic the bat using some correlated walk. If you look at the angle, the heading angle at the moment where the bats take off from point A, let's say to point B while performing shortcuts, you can see that they are distributed very different from a uniform distribution and they're narrow around zero and again they cannot be explained, this distribution cannot be explained by some kind of correlated random walk movement, so the bats seem to know where they're going from the moment that they start. We also detected what we call long cuts, what do I mean by a long cut, it's a term we invented, so if you look at this bat, so everything it did before is within this home range, what we call home range down here and then it performs this long exploratory flight in blue and reaching very large distances far from wherever it has ever been and it is also able to return to in this case a fruit tree and some cases the colony from these very large distances outside its familiar home range and once again we detected something like a hundred such long cuts and once again these long cuts are very straight, not as straight as short cuts but much straighter than any exploratory behavior or any model and once again the animals know where they're going at the moment when they take off. What about the ontogeny of the ability to perform short cuts and long cuts, so there seems to be no difference between night one outside and over time, so what we see here are days outside and what's the rate of performing short cuts and long cuts and this is just noise, I don't think there's any pattern here, the bats can do it from day one outside, by the way, day one outside is not, they're not, it's not day one in their life, usually they're around four weeks when they start, sorry, around eight weeks old when they start flying outside but from that day onwards there's no increase in the rate of short cuts or long cuts, however there is an increase in the length of the short cuts and the long cuts and I think this makes perfect sense, the ability is there but when you map more and more of the region then you can perform longer short cuts and long cuts. What sensory system, which sensory modality is being used to perform this navigation is the next question I'm about to answer, so it has been already proposed by Asaf Tsar who was a study of Fran who we'll talk later on today that these bats use, these same fruit bats use vision when navigating in rural areas so to test this hypothesis we performed several things in order to test this hypothesis and I'll bring you a few points that I think suggest that the bats are indeed using vision once again I remind you many people think that bats are blind so in this case not only they're not blind they probably rely on vision for navigation and I believe that many bats probably do this. Here are a few points that I think suggest visual based navigation and visual based map so what you see here is a correlation between the distance of the shortcut and the altitude of the bat so when the bats return from larger distances or perform longer shortcuts they will elevate, they will ascend to higher altitudes before performing the return if you're using a magnetic sense or I don't know about the fraction we can argue about it later probably it doesn't make much sense to ascend. Why should they ascend? So I will show you now so most of the time these bats are flying very low less than 10 meters this is an urban environment and here is what you see when you fly at these low altitudes so I'm showing you drone images taken by drones flying at the same altitude of the bats at the same locations 11 locations flown by our bats this is what you see okay so basically can you see or should I turn off the light? Turn off the light? Okay I think I know how to do that Okay apparently I know so essentially you see nothing when you fly at these heights I mean there are no visual landmarks that can be used unless you're doing some kind of a template matching algorithm okay so what we would expect is that the height of the bats the height the bats ascend to before performing a long cut and short cut will be correlated to the height of the buildings around them right and that's what we find okay we find a significant positive correlation between the height of the buildings around the bats and the height they ascend to when performing these short cut and long cuts what do you gain by surpassing the buildings I'll show you in a second exactly what you gain so this is a bat this is a drone mimicking a bat moving from its forging grounds which are down there to the height it ascended to before so this is the height it ascended to over here before returning to the colony and once you reach the height this is the height you only have to pass the buildings this is what you can now see okay so there's a huge urban horizon with a lot of landmarks I immediately recognize them of course you do not but I can point out the direction of the colony just by using these distal urban landmarks we also quantify this we took a drone to a 20 location where the bats ascended to we recognized several salient landmarks that we guessed could be meaningful and indeed in all cases the bats by ascending you can see many more of these landmarks than you could if you did not ascend last point I want to make on this story is related to individuality which is something that was not yet much mentioned in this meeting and that is the fact that different bats sometimes people would call it personalities have simple personalities or behavioral traits some people would say and we wondered how can this influence their navigation and what do I mean by personality in terms of navigation so here you can see three individual bats and all of them the tracks of all of them on day 80 outside are shown so there's no difference in the time they spent outside look at this individual how exploratory it is I think it's the same one I showed you the movie of at the beginning the beginning and look at this individual which is our least exploratory but basically all it did is to leave the colony fly to nearby tree and come back so we were guessing that if you indeed then here we quantify this is just the home range sorted according to individuals and we were guessing that if you indeed have to map the environment this will also affect your ability to navigate and in order to test this we performed the translocation experiment which is very common in the navigation literature in which you take the animal to a new location and you release it where it has hopefully never been in our case since we knew exactly where these animals have been before we could actually validate that we release them in a point that they have never been to and it did not have to be very far so we took them only five kilometers away but we knew they have never been there and indeed look at this exploratory bat released five kilometers from the colony by the way all bats have been released in two points so here on the coast and here in the center of the city and look how this exploratory bats how easily it homes back to the colony in blue while this individual coral was released this is the least exploratory individual released here on the coast she clearly had no idea where she's going right she flies straight into the Mediterranean Sea and then at some point she realizes that something is wrong heads back and then returns by the way I cannot really explain how she returns she must have somehow used the landmarks while flying over here to assess the direction the azimuth of the colony if you're asking yourself and this is indeed repeatable I didn't show you the graph but you can see a nice correlation between your exploratory tendency and your straightness of homing and if you're asking yourself what happened over here at this point so now I overlay the altitude of the bat on the same trajectory and you can see that she's climbing this is only 50 meters but there's a cliff over here and if I interpret what's going on in her mind so she's flying and at some point she looks back and she realizes oops the city is over there and this is this point and then she turns and descends and goes back to the colony so what do I think is going on I already hinted on this before I think it's a visual map I think it's based on triangulation of salient landmarks, buildings but also maybe junctions maybe highways as well and the bats can use this in order to estimate their azimuth to the desired azimuth sorry, towards their destination maybe also distance I don't know we're testing I mean something about the distance they know for sure here's just a nice example to show to prove that this has to be an angle a point of view independent base representation so here's a bat and this happens very often so the home range of the bats is over here almost the city is here so the landmarks are down here and now this bat finds itself south to these landmarks for the first time and this happens a lot even in that movie that I showed you so suddenly the arrangement of the landmarks reverses you can see this is what she's seen before and this is what she sees it how she sees it now everything is reversed and the distances the angular distances between them of course have changed completely because she's far down here and still she finds her way or it finds its way back to a tree or a column or the colony and without any problems the last thing I think I want to show you because I'm running out of time is we're trying now to quantify how can they use these visual landmarks to navigate I gave you some kind of intuition but we're trying to quantify it more accurately for that we're using bats that navigate in these rural areas because there's simply there's I think less information than in the city where it's very very difficult and what we're doing is what we did actually is to fly a drone in the exact flight trajectory of the bat collecting visual information from the point of view of the bat so this is the trajectory the bat this is something like a 20 kilometer trajectory the bat leaves its roost and flies to a tree 20 kilometers away this is the information available so it's flying towards the coast in Israel and it of course at no point can see the tree which is 20 kilometers down there all it can see is these background lights and what we've done is to run this through a machine learning through an artificial neural network a convolutional network that the task of the network is to receive an image and to as an output to point towards the direction the right direction the direction of the target and indeed the network can learn this very very precisely so with very little angular error but we did even more than that so after collecting the data along the trajectory of the bat we moved several kilometers away from the trajectory to these blue points up to five kilometers away and we collected visual information in these points on a completely different night by the way and now we're asking ourselves can the same network now predict the direction of the target and indeed it can so you can see this is just an illustration because it's something again we're just now trying to write it up but let's say the bat drifted or for some reason found itself a few kilometers off a route and it is now looking in this direction this is the image that it is seeing we put this as we insert this to the network as an input and the output points towards the target with an accuracy of several degrees so this could explain how they navigate as we discussed two days ago yesterday I think these neural networks are black boxes so what exactly is it using is something we will have to work hard to understand the final point that I want to make is that I want to say something about the ecology of the animal I've shown you two very different types of navigation okay so I showed you these rural bats I've you've seen a few examples of them that fly in straight lines night after night following the center directories very boring navigation if you wish and I showed you how urban bats move and this is what you see down here and it's very very different okay so they are much more diverse in their behavior some people would say they are bar hoppers right they move between trees on consecutive nights they move within the same night they will move from one tree to the other so this notion that their use of navigation is very very different made us think that maybe something in their brain has also adjusted to these different types of navigation I'm not talking evolutionarily I'm talking about behavioral and neural plasticity in this case many of you maybe are familiar with the taxi driver findings in London so what we do very briefly is we MRI the animals and we compare city and rural bats and we also compare rural bats that are brought into the city and we compare their brains before and after several weeks in an urban like environment and indeed we find the differences that could reflect neural plasticity in areas that are related to navigations such as the hippocampus and internal cortex and in sensory areas and also in parts of the limbic system such as the amygdala I think I don't have time so I will skip the last story and I will just ah I want to invite all of you and invitations will come somehow hopefully you will be on some email list that will receive this invitation in March 2020 in the Weizmann Institute in Israel we're organizing a big conference on active sensing that will include people from many different disciplines working on active sensing whisking, echolocation, electro-location, active vision and robotics as well I think many people in this crowd are extremely relevant for this meeting and now I want to thank a lot of students who collected all of this data and collaborators and funding agencies and that's it, thank you very much