0:01 Hi, I'm Laura. And I'm here today with the first part of our swarm series. And I'm a PhD student at the University of Bristol as is my colleague, Julian, who will do part two. And we are both on the robotics PhD, and it's called first go on, there's a link to it in the document. And so I'm just gonna start off 0:24 with some aims. So today, I'm here to introduce you to swarm engineering, why it's useful and why I think it's so cool. And hopefully, you'll think that's cool, too. And I want you to u nderstand where it comes from algorithms are used in real life. 1:14 Just gonna do some definitions that might be useful throughout this course to help you understand what we're talking about. So I've said the word swarm a few times. And if you don't re ally understand what a swarm is, it's just a large group of individuals. And swarming is when a large group of individuals display a collective behavior. So whenever they do something interesting together, for example, and in flocking individuals are birds and the behavior is high they move. An individual can also be called an agent. We use this word quite a lot when we're talking about it. And this can be a human, a fish, an insect or a robot. It's just one of the individuals in the swarm. And swarm engine ering is when we try to design individuals so they can 2:00 could be in a simulation, it could be a robot that could be particles, and to actually have a collective behavior. So maybe we want to have drones that fly in the sky that way birds do. Maybe w e want to have tiny robots that move around the way ants do. And that's just what we mean by swarm engineering. Why do we use swarms rather than just other types of engineering or other types of robots, and the three main things are robust, flexible, and s calable. And if you imagine these little dots that I've drawn are little insects moving around. So with robust why it's more robust than something else is because there isn't a leader. So every every individual in this form, moves around. If you've ever tr ied to kill ants that are coming to your picnic or come into your kitchen. You notice that even if you kill a bunch of them in the middle, they continue to keep moving. So that's what we mean by robust stick and stand different things. 3:00 Flexible mea ns that it will adapt to the task at hand. So this little red.is has gotten in the way of insects moving around. And some robots or some other forms of engineering that might break the system. Because there's lots of small robots, they can actually move ar ound themselves. And they can be flexible to the situation on the tasks at hand. and scalable means it doesn't really matter if more or less are added to the to the swarm. So they should continue to have the same behavior, even if hundreds more added, even if there's only 10 if someone leaves so for example, in the birds fly around in the sky, if a bird leaves or moves away, it doesn't actually matter. Oh, and one other thing for robust is, if you think about the example of the birds or the fish again, and why they move, and they're actually more robust from attacks from predators because they're a new group, if you think about one bird or fish moving around by itself, a bigger bird or bigger Fish might attack them, where as they want whenever they're less l ikely to whenever there's a big group of them, or the challenges with swarm engineering? Well, I've been using a bird example quite a lot. So if you look at these birds flying around, we can see what the end behaviors and behaviors, we want them to move in a flock, and to move like this and to move like that. But what are the actual birds, the individual birds doing, so that the group has this behavior. So that's quite challenging to actually think about it, and to actually try and design the individuals to have the group behavior. 4:35 The methods that we use for this our bio inspiration, which literally just means we look at biology, so we might look at animals, insects, plants, humans, and see what they're doing and try to work it work. Another way is crowdsourcing. So crowdsourcing is whenever you try to get lots of people to help you with something and quite often it's about putting a game or an idea on the internet. So what we've done is there was a game called nano de Which was hard to design and ho w to design particles that could do trunk cancer treatment in the body, and we put it online so everyone could play it. So we could see how they made different particles, it was really interesting. And we were able to analyze the results, we've got thousan ds of people playing an increasingly popular method is machine learning. So that's when you use algorithms and coding to try and crack the behavior. And you can do this by running lots of simulations, which is when you have a program that tests lots of thi ngs out for you. And usually some combination of the three works well whenever we're trying to do swarm engineering. Now I'm going to talk to you about some actual algorithms. So I'm obviously going to start with flocking because I've mentioned it so many times. And if you look at these and pretend that they're birds, excuse my drawing, and every bird has a neighborhood, every bird can speak to the birds around it. So the one in the middle of the circle, pretend that's its neighborhood and it can speak To t he four birds around it. Why I really like flocking is because it has three really simple rules actually. So that complex behavior can be broken down into three really simple rules. They are separation cohesion and alignment. on higher I think of this is, cohesion means stay close to your neighbors separation means not too close that you're bumping into them. And alignment means point towards the center. So with these three rules when they were implemented in some drones, and with as little as 12, they actu ally got to behave in the flocking and behavior and other birds actually joined in which is really interesting. And floggings quite a cool algorithm. And you actually probably all seen us so it's quite useful for game design. So if you've seen the original Lion King the cartoon, whenever the wildebeest run that was done through flocking, if you've seen Batman, hire the birds fine. Batman is done with the flocking algorithm. Another one is called particle swarm optimization. And this is seen in birds and fis h. So I always think about this as how birds decide which trees to move to. And to get food. So how it works is they try to predict where the particles so the bird or the or the fish is going to move next. And and in birds, how it's done is if you think of these green dots as food sources, so the birds sitting out the food sources cry out in a certain way. And the birds in the middle, decide where it's best to go to. And, and the optimization is when we actually try to work this out and predict this. It's q uite interesting. There's different methods that can be used. So the particles can be brave, which means they move to new locations, and they try to go to the unexplored, which if you look at this would actually be quite useful because there's one big gree n.of food that hasn't been explored yet. It can be conservative, which means they only move to sources that they know are good, or they can swarm Which just means they spread out as best as possible. So this has been tested as to what's the best method. An d the best method is if a if it's a swarm, and takes a mixed approach. So some of the particles in the swarm are brave, some are conservative and some swarm. That's how they find that's how they cover best. And then particle swarm optimization is used in a lot of really interesting real world applications. So one is traffic accident forecasting. So if we can count the cars as particles, and then we can and model where they're going to move to next, we can maybe work out if there's going to be accidents on t he motorway, or if cars are going to go in too fast, and they're going to drive the traffic lights, which is really interesting. It's also used in medical imaging analysis. And so that is counting the light particles that are used in certain scans in the b ody and, and counting the light particles as particles in the particle swarm optimization. I'm predicting more They shouldn't move. So if I predict, say, Hi something, you should move through a certain part of the body, if it doesn't move that way, then pe rhaps there's a tumor or an obstruction or something there that needs to be looked into. 9:11 And the final algorithm I'm going to talk to you about is called ant colony optimization. And and this is how ants move through their colonies. And it's quite interesting. It's hard to map the best way to move an answer very interesting. In general, there's quite a lot of swarm algorithms that are based on us. And so as the ants move through the colony, they secrete a pheromone on their own skin smell, and they leave it behind and as dance moves through, you can see in the second image, both sides are filled up. But what happens is in the shorter paths, the pheromone is stronger. So that's the way the answers start to follow. So all the answers start to fill in a nd on the shortest path. So why this is really interesting is it's used in high Wi Fi networks are done so and the Wi Fi and Engineers kind of do a test of all the network. So they map out the shortest route to your house. But they also use it to totally m odify the network. So ants will know they're calling me really well. And they'll know what happens if a bit gets obstructed. That's the same way with Wi Fi networks just to make sure that you get continuous service of one of the wires or cables is done. An d it'll know the best path kick next. That's everything from me. And there is a homework, which is to look into a few other m of the swarm algorithms. So one of the examples I would say with maybe to look at clustering or decision making, but there is a Wi kipedia page with a really good list of different form algorithms. And if you're feeling creative, and a good way to learn about these is to actually build an infographic. So I've given you an example for flocking, and you can build another one. And Part t wo is also available where Julian is going to be talking about swarm robotics. I really hope you've enjoyed this. Thank you. Transcribed by https://otter.ai