Robots that fly … and cooperate | Vijay Kumar
Translator: Liridon Shala Reviewer: Spartak Ferrollari Good day. I'm here today to talk around the flying and autonomous beach balls. No, agile air robots like this. I want to tell you something about the challenges in building these and some amazing opportunities for the application of this technology. So these robots are related to remotely commanded aircraft However, the tools you see here are great. They weigh several tons, are not by all means agile. They are not even autonomous. In fact, most of these cars operated by flight crews which may involve many pilots sensor operators and mission coordinators. What we are interested in is the development of robots such as this – and here are two more pictures – of robots that you can buy in the store.
So these are four-rotor helicopters gave are approximately one meter large and weigh several kilograms. And so we added sensors and processors to them, and these robots can fly indoors without GPS (Global Positioning System). The robot I am holding in my hand is this, and was created by two students, Alex and Daniel. So this one weighs less se 50 gr? Consumes about 15 watts of electricity. And as you can see, has about 20 cm in diameter. So let me give you a quick guide how these robots work. So there are four rotors. If you spin these rotors at the same speed, the robot hovers. If you increase the speed of each of these rotors, then the robot flies up, accelerates. Of course, if the robot was tilted horizontally then it will accelerate in that direction. So to tilt it, there are one or two ways to do it. So in this picture you can see that the fourth rotor is spinning faster and the second rotor is spinning more slowly. And whenever it happens is a moment that makes the robot spin.
On the other side, if it increases the speed of the third rotor and reduces the speed of the first rotor then the robot will advance forward. And then finally, if you turn it on the opposite side of the rotors faster than the other connection, then the robot goes around the vertical axis. So a onboard processor observes what moves need to be executed and combines these movements and finds out what command should be given to the engines 600 times per second. This is basically how these things work. So one of the advantages of this project is that when you reduce the amount of this thing, the robot naturally becomes agile. So here R is the robot length characteristic. It is actually half the diameter.
and there are many physical parameters which vary when you reject R. What is most important is inertia or resistance to motion So it turns out to be, that inertia, which regulates motion in action, varies as the fourth power of R. So the smaller you make R, the most dramatic becomes the reduction of inertia. And as a result, the angular acceleration, written in Greek alphabet alpha, goes over an R. It is proportionally opposite to R. The smaller you make it the faster it can turn. So that should be clear in this video. At the end you can see a robot which performs in a 360 degree turn in less than half a second Some returns, even in less time. So here is the process on board are receiving ratings from accelerometers and board gyroscopes and calculated, as mentioned above, commands at 600 times per second to stabilize this robot. So on the left, you can see Daniel looks thrown this robot into the air. And I show how strong his control is. It doesn't matter how you shoot it, the robot turns and comes back to him.
So why build robots like these? Well then, such robots have many applications. You can send them to buildings like these as the first response in search of intruders perhaps to search for biochemical leaks, gas leak. You can use them for applications like constructions. Here you have robots pulling beams, columns and assembling cubic structures. I will tell you something more about this. Robots can be used for cargo transportation. So one of the problems with these little robots is their carrying capacity.
So maybe you might want to have some robots for the transport of useful cargo. This is a photo of an experiment we did recently – actually not so late – in Sendai immediately after the earthquake. So such robots can be sent to closed objects in assessing damage after a natural disaster, or be sent to reactive buildings to design radioactivity levels. So a basic problem that robots must solve, if they are autonomous is to make the right choice how to get from point A to point B. So this is a bit challenging because the dynamics of the robot is very complex.
In fact, they live in a 12-dimensional space So we use a little trick. We get this 12 dimensional curved line and we transform it in a flat four-dimensional space. And this four dimensional space consists of X, Y, Z and then another curve. And what the robot does is his plan which we call the minimal jump trajectory. So to remind you of physics, you have position, derivatives, speed, then acceleration, and then comes the oscillation and then comes the jump. So this robot minimizes cracking.
So what effectively does is producing a smooth and pleasant motion And they made him run away from obstacles. So these minimal trajectory jumps in flat spaces are transformed again in this complex 12-dimensional space, which robots have to do for control and then execution. So let me show you some examples of the minimum of what these jump trajectories can do. And in the first video, you will see the robot going from point A to point B. through an intermediate point. So the robot is definitely capable for the execution of any curve trajectory. So these are circular trajectories where the robot pushes about 2 G. (G: gravitational force) Here we have a camera mounted on top which captures moving images which shows where the robot is at 100 times per second. also tells the robot where the obstacles are.
Obstacles can also move. So here we have Daniel throwing a circle in the air, while the robot is calculating the position of the circle and is trying to find the best way to get through the circle. So as an academic, we are trained to always be able to jump through the circles to raise funds for our labs, and we make our robots do it. (Applause) Another thing robots can do is to recall parts of the trajectory who has taught them or who has been pre-programmed. Here you can see a robot performing a movement to gain momentum and then changes its orientation to redirect. So it has to do with this as this space in the window is only slightly larger than the width of the robot. So just like a diver stands on a springboard and then jumps to gain momentum, then do this pirouette and a somersault over and then gracefully returns to the position, this robot more or less did it.
So they know how to combine some of the parts of the trajectory to make these tasks quite difficult. So I want to change the subject. One of the disadvantages of this small robot is its size. And I told you before that we want to use a lot of robots to exceed the size limits. So a problem is how to coordinate many of these robots? So here we have seen nature. I want to show you a video of the desert ants of the Aphaenogaster family in the laboratory of Professor Stephen Pratt holding an object. It's actually a piece of fig. In fact you get every object stained with fig juice and the ants will lead them to the nest.
So these ants do not have any central coordinator. They feel the presence of neighbors. There is no clear communication. But because they feel their neighbors and because they feel objects, they have implicit coordination within the group. So this is the kind of coordination that we want our robots to have. So when we have a robot which is surrounded by neighbors – and let's look at robot I and robot J – what we want robots to do is to monitor the distance between them when they fly in formation.
And then you want to be safe that this distance is at acceptable levels. Again the robots monitor this error and calculate control commands 100 times per second, which then translates to engine commands 600 times per second. So this should definitely be done in a decentralized manner. Again, if you have lots and lots of robots, it is impossible to coordinate all this information centrally fast enough for the robots to perform the task. Plus robots have to base their movements yourself in local information, what he feels from their neighbors. And then finally, we insist that robots do not recognize who are their neighbors. So this is what we call anonymity. So what I want to show further is a video of 20 small robots flying in formation. They are observing the position of their neighbors. They are guarding the lineup. Formations may vary. They can be planar formations, they can be three-dimensional formations. As you can see here, they collapse from a three-dimensional formation into planar formation. And to fly through obstacles they can adapt the formation in flight. So once again, these robots go very close to each other.
As you can see in this picture, the eighth of the flight, they move within inches of each other. And despite the aerodynamic interactions of propeller blades, they are able to maintain stable flight. (Applause) So once you know how to fly in formation, you can actually raise objects in interaction. So that just shows that we can double, triple, or quadruple robot force just by joining them with the neighbors, as you can see here.
One of the disadvantages of doing it is how we increase the dimension of objects – so the number of robots increases by keeping the same, basically increase inertia, and so you pay a price, they are not so agile. But you win in terms of carrying capacity. Another app I want to show you – again, this is in our lab. This work was performed by Quentin Lindesy who is a graduate student. So this algorithm basically tells these robots how to build in autonomy cubic structures from different elements.
So his algorithm tells the robot which part to take, where and when to place it. So in this video you can see – accelerated 10, 14 times – you can see three different structures built by these robots. And yet, everything is independent, and all that Quentin has to do is to give them a copy drawing what they want to build. So all these experiments you've seen so far, all these demonstrations, are made with the help of motion capture systems.
So what happens when you leave your lab and you go out into the real world? And what if there is no GPS? (GPS: Global Positioning System) So this robot it is actually equipped with a camera and an H detection laser, laser scanner. and uses these sensors to develop a map of the surrounding environment. The map consists of many features – like doors, windows, people, furniture – and then calculates where his position is in terms of features. So there is no global coordination system. The coordinate system is based on robots, where it is and what it is looking for. And it navigates about those features. So I want to show you a clip of algorithms developed by Frank Shen and Professor Nathan Michael who displays this robot entering a building for the first time creating this map in flight. So the robot determines what the characteristics are. Develops the map. Calculate where it is found in relation to the characteristics of the environment and evaluates its location 100 times per second allowing us to control algorithms for whom you described earlier.
So this robot is actually being commanded at a distance from Frank. But the robot can count on its own where to go. Suppose I had to send him to a building and I had no idea what that building looked like, I can ask this robot to go, create a map go back and show me the features of that building. So here, the robot is not just solving the problem, how to get from point A to point B on this map, but he determines which is the best point B at any time. So I actually know where to go to search for places about which there is less information. And that is how this map is completed. So I want to leave you with a recent application. And there are many ways this application can be used.
I am a professor, and we are passionate about education. Such robots can really change the way of education in primary to secondary schools But we're in Southern California, near Los Angeles so I have to finish with something focused on fun. I want to end with a music video. I want to introduce the creators, Alex and Daniel, who created this video. (Applause) So before I play this video, I want to tell you that they created it in the last three days after receiving a call from Chris. And robots that play in video are completely independent. You will see nine robots playing six instruments. And of course, this is done exclusively for TED 2012. Let's look at it. (Music) (Applause).