Translation: Norbert Langkau Editing: Daniel Hoffmann When is a machine athletic? We'll get the concept of machine athletics and the research needed with the help of these aircraft, the quadrocopters or “quads” for short, demonstrate. Quads have been around for a long time. You are so popular today because they are mechanically simple. By controlling the four propellers these machines can roll, tip, yaw and accelerate in all common directions. Also on board are a battery, a computer, various sensors and radios. Quads are extremely manoeuvrable, but this maneuverability comes at a price. They are inherently unstable, and need a kind of automatic feedback control for flying. How did he just do that? Cameras on the ceiling and a laptop serve as an indoor GPS. This allows objects in the room, that have such reflectors, localize. The data is sent to another laptop sent with estimation and control algorithms and from there to the quad, which also executes such algorithms. We mainly research algorithms. You are the magic that brings these machines to life. So how do you design algorithms that make a machine so sporty? We use the so-called model-based approach.

First we form the physical behavior of the machines in a mathematical model. Then we analyze these models using a sub-area of ​​mathematics, control theory, which then automatically supplies the algorithms for their control. An example is this floating quad: First we have the dynamics mapped in several differential equations, which we then change with the help of control theory and so get algorithms to stabilize the quad. I want to show you the strengths of this approach. Let's assume that the quad is supposed to hover in addition to it still balance this stick. With practice that is a manageable task for people, where we to our advantage two feet on the floor and have our very skilled hands.

It'll be a little more difficult with just one foot on the ground and without hands. The stick has a reflector at the tip, so that it can be localized in space. (Applause) You can see the quad through small adjustments keeping the staff in balance. How did we design the algorithm for it? We have the mathematical model of the stick added to that of the quad. The rule theory can again be applied to the connected quad-rod system, in order to get algorithms for the control. You see the system is stable and even if I give him little nudges it goes back into a nicely balanced position. We can do the model too expand to a desired position. With this pointer made of reflectors I can point to where I put the quad in space at a fixed distance from me.

The key to this acrobatics are algorithms, which with the help of mathematical models and control theory were developed. Now the quad is supposed to come back here and drop the staff. Next I want to show how important it is understanding physical models and whose connections are in the physical world. You can see that he is losing height when I put this glass of water on it. Unlike the balance stick, I have here the model of the glass is not inserted into the system.

The system actually doesn't even know that the water glass is there. As before, I can use the pointer to tell the quad to where in the room I want him. (Applause) At this point you ask yourself: "Why doesn't the water flow out of the glass?" For two reasons: On the one hand, gravity works the same for all objects; on the other hand, all propellers point in the same direction as the glass, upwards. Both together cause that the side forces on the glass are small and are predominantly aerodynamic effects, which can be neglected at these speeds. Therefore you do not need to insert the model of the glass. Nothing is spilled, no matter what the quad does. (Applause) The lesson we learn from this is that some high achievements are easier than others, and that physics in the problem tells you what is easy and what is difficult. Here, for example, it is easy to carry a glass of water, but difficult to balance the stick. We've all heard of athletes who also deliver masterpieces injured. Can a machine do that too? even if it is badly damaged? Traditionally you need at least four permanently attached, powered propeller pairs to fly, because there are four degrees of freedom: Roll, tilt, yaw and accelerate.

Hexacopter and Octocopter, with six and eight propellers, are redundant, but quadcopters are much more popular, because they are the smallest number Fixed, powered propeller pairs have: four. Or not? The mathematical model of this machine with only two intact propellers shows that there is an unconventional way of flying them. We renounce yaw, but control rolling, tilting and acceleration by algorithms that we get from the evaluation of this new arrangement. Mathematical models tell us exactly when and why this is possible. Here this knowledge allows us novel machine architectures or design clever algorithms that work with destruction just as well treat like athletes, instead of having to build machines with redundancy. We catch our breath when we see like a high diver does a somersault in the water, or a pole vaulter bends in the air, and quickly gets close to the ground.

Will the high diver dip smoothly? Will the pole vaulter make the landing? This quad is supposed to be one do a triple somersault and again arrive where it started. This maneuver happens so fast that we cannot feed back during the movement – there is simply not enough time for that. Instead, the quad can fly the maneuver blind, look at the result and then learn from this information so that the next role will be better. As for the jumpers, that the maneuver only learned through repeated practice and can be carried out optimally. (Applause) Many sports require the ability to hit a moving ball. How do you leave a machine do something that seems effortless to an athlete? (Applause) This quad has a racket mounted on its head, whose clubface is not very large, about the size of an apple.

Every 20 milliseconds, i.e. 50 times per second, we calculate the following: First we calculate where the ball is going. Then how the quad should hit the ball, so that it flies back to the thrower. Third, a trajectory is planned that will guide the quad from its position to its point of service for the ball. Fourth, we only follow this strategy for 20 milliseconds. The process is then repeated for as long as until the quad hits the ball. (Applause) Machines can not only perform such maneuvers alone, but also in a network. These three quads together hold a net in the air. (Applause) You are performing an extremely dynamic one and coordinated maneuvers, to kick the ball back to me. Look – at full expansion, the quads are vertical. (Applause) Indeed, the forces are at full extension about five times the size of a bungee jumper felt at the end of the jump. The necessary algorithms are those for the single quad to throw the ball back, very similar.

So one uses mathematical models, to recalculate the coordinated strategy 50 times per second. So far we've been about machines and their capabilities. What if we are having this machine athletics that of a human pair? This is a normal gesture recognition sensor as used in video games. He recognizes in real time what my body parts are doing right now. Similar to the pointer earlier we can use it as an input medium. Now we have a natural way to control the untamed athleticism of these quads with my gestures.

(Applause) The interaction does not have to remain virtual, it can become physical. Let's take this quad here. He tries to keep his position. If I try to move him away, he'll fight back and moves back to the position it wants to maintain. But we can change this behavior. With mathematical models you can estimate the force that is acting on the quad. With this we can change the laws of physics – of course only insofar as they concern the quad.

Here the quad behaves, as if it were in a viscous liquid. Now we have a method interact closely with the machine. With this I can position this quad so that he did the rest of the demonstration can film from above with the mounted camera. So we can interact with these quads and change the laws of physics. We're having fun with that now.

Next act the quads initially as if they were on Pluto. Over time the force of gravity increases until we're back on earth but we never get there. Okay, let's go. (laugh) (laugh) (Applause) Uff! You are probably thinking: "The guys are having too much fun" and you are probably wondering: "Why do they actually build machine athletes?" It is believed that animals play to develop dexterity and skills. Others think it works socially and ensures group cohesion. We use the analogy between sport and athletics, to develop new algorithms, that the machines can exhaust. How does the speed of machines affect our lives? Like all human inventions and innovations can they improve human life – but they can also be abused. This is not a technical decision – but a social one. With the right choice we create for the future of machines like through athletics in sport with us the best results. Now to the wizards behind the green curtain. They form the Flying Machine Arena research team. (Applause) Federico Augugliaro, Dario Brescianini, Markus Hehn, Sergei Lupashin, Mark Muller and Robin Ritz.

Something will come of them again! Many Thanks. (Applause).