The astounding athletic power of quadcopters
-
0:11 - 0:14So what does it mean for a machine to be athletic?
-
0:14 - 0:18We will demonstrate the concept of machine athleticism
-
0:18 - 0:20and the research to achieve it
-
0:20 - 0:22with the help of these flying machines called quadrocopters,
-
0:22 - 0:24or quads, for short.
-
0:26 - 0:29Quads have been around for a long time.
-
0:29 - 0:30The reason that they're so popular these days
-
0:30 - 0:32is because they're mechanically simple.
-
0:32 - 0:34By controlling the speeds of these four propellers,
-
0:34 - 0:37these machines can roll, pitch, yaw,
-
0:37 - 0:40and accelerate along their common orientation.
-
0:40 - 0:43On board are also a battery, a computer,
-
0:43 - 0:47various sensors and wireless radios.
-
0:47 - 0:52Quads are extremely agile, but this agility comes at a cost.
-
0:52 - 0:55They are inherently unstable, and they need some form
-
0:55 - 0:59of automatic feedback control in order to be able to fly.
-
1:04 - 1:07So, how did it just do that?
-
1:07 - 1:09Cameras on the ceiling and a laptop
-
1:09 - 1:12serve as an indoor global positioning system.
-
1:12 - 1:14It's used to locate objects in the space
-
1:14 - 1:17that have these reflective markers on them.
-
1:17 - 1:19This data is then sent to another laptop
-
1:19 - 1:21that is running estimation and control algorithms,
-
1:21 - 1:23which in turn sends commands to the quad,
-
1:23 - 1:26which is also running estimation and control algorithms.
-
1:30 - 1:32The bulk of our research is algorithms.
-
1:32 - 1:36It's the magic that brings these machines to life.
-
1:36 - 1:38So how does one design the algorithms
-
1:38 - 1:41that create a machine athlete?
-
1:41 - 1:43We use something broadly called model-based design.
-
1:43 - 1:47We first capture the physics with a mathematical model
-
1:47 - 1:49of how the machines behave.
-
1:49 - 1:51We then use a branch of mathematics
-
1:51 - 1:54called control theory to analyze these models
-
1:54 - 1:58and also to synthesize algorithms for controlling them.
-
1:58 - 2:01For example, that's how we can make the quad hover.
-
2:01 - 2:02We first captured the dynamics
-
2:02 - 2:04with a set of differential equations.
-
2:04 - 2:07We then manipulate these equations with the help
-
2:07 - 2:11of control theory to create algorithms that stabilize the quad.
-
2:11 - 2:14Let me demonstrate the strength of this approach.
-
2:17 - 2:20Suppose that we want this quad to not only hover
-
2:20 - 2:23but to also balance this pole.
-
2:23 - 2:24With a little bit of practice,
-
2:24 - 2:27it's pretty straightforward for a human being to do this,
-
2:27 - 2:29although we do have the advantage of having
-
2:29 - 2:30two feet on the ground
-
2:30 - 2:33and the use of our very versatile hands.
-
2:33 - 2:35It becomes a little bit more difficult
-
2:35 - 2:38when I only have one foot on the ground
-
2:38 - 2:40and when I don't use my hands.
-
2:40 - 2:43Notice how this pole has a reflective marker on top,
-
2:43 - 2:47which means that it can be located in the space.
-
2:53 - 2:59(Applause)
-
2:59 - 3:02You can notice that this quad is making fine adjustments
-
3:02 - 3:04to keep the pole balanced.
-
3:04 - 3:07How did we design the algorithms to do this?
-
3:07 - 3:09We added the mathematical model of the pole
-
3:09 - 3:11to that of the quad.
-
3:11 - 3:14Once we have a model of the combined quad-pole system,
-
3:14 - 3:19we can use control theory to create algorithms for controlling it.
-
3:19 - 3:20Here, you see that it's stable,
-
3:20 - 3:23and even if I give it little nudges,
-
3:23 - 3:28it goes back to the nice, balanced position.
-
3:28 - 3:30We can also augment the model to include
-
3:30 - 3:32where we want the quad to be in space.
-
3:32 - 3:35Using this pointer, made out of reflective markers,
-
3:35 - 3:38I can point to where I want the quad to be in space
-
3:38 - 3:41a fixed distance away from me.
-
3:56 - 3:59The key to these acrobatic maneuvers is algorithms,
-
3:59 - 4:01designed with the help of mathematical models
-
4:01 - 4:03and control theory.
-
4:03 - 4:05Let's tell the quad to come back here
-
4:05 - 4:07and let the pole drop,
-
4:07 - 4:09and I will next demonstrate the importance
-
4:09 - 4:11of understanding physical models
-
4:11 - 4:15and the workings of the physical world.
-
4:25 - 4:27Notice how the quad lost altitude
-
4:27 - 4:29when I put this glass of water on it.
-
4:29 - 4:32Unlike the balancing pole, I did not include
-
4:32 - 4:35the mathematical model of the glass in the system.
-
4:35 - 4:38In fact, the system doesn't even know that the glass of water is there.
-
4:38 - 4:41Like before, I could use the pointer to tell the quad
-
4:41 - 4:43where I want it to be in space.
-
4:43 - 4:53(Applause)
-
4:53 - 4:55Okay, you should be asking yourself,
-
4:55 - 4:58why doesn't the water fall out of the glass?
-
4:58 - 5:01Two facts: The first is that gravity acts
-
5:01 - 5:03on all objects in the same way.
-
5:03 - 5:06The second is that the propellers are all pointing
-
5:06 - 5:09in the same direction of the glass, pointing up.
-
5:09 - 5:11You put these two things together, the net result
-
5:11 - 5:13is that all side forces on the glass are small
-
5:13 - 5:16and are mainly dominated by aerodynamic effects,
-
5:16 - 5:20which as these speeds are negligible.
-
5:23 - 5:25And that's why you don't need to model the glass.
-
5:25 - 5:29It naturally doesn't spill no matter what the quad does.
-
5:39 - 5:46(Applause)
-
5:46 - 5:50The lesson here is that some high-performance tasks
-
5:50 - 5:51are easier than others,
-
5:51 - 5:53and that understanding the physics of the problem
-
5:53 - 5:56tells you which ones are easy and which ones are hard.
-
5:56 - 5:58In this instance, carrying a glass of water is easy.
-
5:58 - 6:02Balancing a pole is hard.
-
6:02 - 6:04We've all heard stories of athletes
-
6:04 - 6:06performing feats while physically injured.
-
6:06 - 6:08Can a machine also perform
-
6:08 - 6:11with extreme physical damage?
-
6:11 - 6:12Conventional wisdom says that you need
-
6:12 - 6:16at least four fixed motor propeller pairs in order to fly,
-
6:16 - 6:18because there are four degrees of freedom to control:
-
6:18 - 6:21roll, pitch, yaw and acceleration.
-
6:21 - 6:24Hexacopters and octocopters, with six and eight propellers,
-
6:24 - 6:26can provide redundancy,
-
6:26 - 6:28but quadrocopters are much more popular
-
6:28 - 6:30because they have the minimum number
-
6:30 - 6:32of fixed motor propeller pairs: four.
-
6:32 - 6:34Or do they?
-
6:49 - 6:52If we analyze the mathematical model of this machine
-
6:52 - 6:54with only two working propellers,
-
6:54 - 7:01we discover that there's an unconventional way to fly it.
-
7:08 - 7:10We relinquish control of yaw,
-
7:10 - 7:13but roll, pitch and acceleration can still be controlled
-
7:13 - 7:18with algorithms that exploit this new configuration.
-
7:22 - 7:24Mathematical models tell us exactly when
-
7:24 - 7:26and why this is possible.
-
7:26 - 7:29In this instance, this knowledge allows us to design
-
7:29 - 7:31novel machine architectures
-
7:31 - 7:35or to design clever algorithms that gracefully handle damage,
-
7:35 - 7:37just like human athletes do,
-
7:37 - 7:41instead of building machines with redundancy.
-
7:41 - 7:43We can't help but hold our breath when we watch
-
7:43 - 7:45a diver somersaulting into the water,
-
7:45 - 7:47or when a vaulter is twisting in the air,
-
7:47 - 7:49the ground fast approaching.
-
7:49 - 7:51Will the diver be able to pull off a rip entry?
-
7:51 - 7:53Will the vaulter stick the landing?
-
7:53 - 7:55Suppose we want this quad here
-
7:55 - 7:57to perform a triple flip and finish off
-
7:57 - 8:00at the exact same spot that it started.
-
8:00 - 8:02This maneuver is going to happen so quickly
-
8:02 - 8:06that we can't use position feedback to correct the motion during execution.
-
8:06 - 8:08There simply isn't enough time.
-
8:08 - 8:11Instead, what the quad can do is perform the maneuver blindly,
-
8:11 - 8:14observe how it finishes the maneuver,
-
8:14 - 8:16and then use that information to modify its behavior
-
8:16 - 8:18so that the next flip is better.
-
8:18 - 8:20Similar to the diver and the vaulter,
-
8:20 - 8:22it is only through repeated practice
-
8:22 - 8:24that the maneuver can be learned and executed
-
8:24 - 8:26to the highest standard.
-
8:34 - 8:39(Applause)
-
8:39 - 8:43Striking a moving ball is a necessary skill in many sports.
-
8:43 - 8:44How do we make a machine do
-
8:44 - 8:48what an athlete does seemingly without effort?
-
9:04 - 9:11(Applause)
-
9:11 - 9:13This quad has a racket strapped onto its head
-
9:13 - 9:17with a sweet spot roughly the size of an apple, so not too large.
-
9:17 - 9:20The following calculations are made every 20 milliseconds,
-
9:20 - 9:22or 50 times per second.
-
9:22 - 9:24We first figure out where the ball is going.
-
9:24 - 9:27We then next calculate how the quad should hit the ball
-
9:27 - 9:30so that it flies to where it was thrown from.
-
9:30 - 9:34Third, a trajectory is planned that carries the quad
-
9:34 - 9:37from its current state to the impact point with the ball.
-
9:37 - 9:41Fourth, we only execute 20 milliseconds' worth of that strategy.
-
9:41 - 9:44Twenty milliseconds later, the whole process is repeated
-
9:44 - 9:46until the quad strikes the ball.
-
9:56 - 9:58(Applause)
-
9:58 - 10:02Machines can not only perform dynamic maneuvers on their own,
-
10:02 - 10:03they can do it collectively.
-
10:03 - 10:07These three quads are cooperatively carrying a sky net.
-
10:17 - 10:22(Applause)
-
10:22 - 10:24They perform an extremely dynamic
-
10:24 - 10:26and collective maneuver
-
10:26 - 10:28to launch the ball back to me.
-
10:28 - 10:32Notice that, at full extension, these quads are vertical.
-
10:36 - 10:38(Applause)
-
10:38 - 10:41In fact, when fully extended,
-
10:41 - 10:43this is roughly five times greater than what a bungee jumper feels
-
10:43 - 10:48at the end of their launch.
-
10:51 - 10:54The algorithms to do this are very similar
-
10:54 - 10:57to what the single quad used to hit the ball back to me.
-
10:57 - 11:00Mathematical models are used to continuously re-plan
-
11:00 - 11:04a cooperative strategy 50 times per second.
-
11:04 - 11:06Everything we have seen so far has been
-
11:06 - 11:09about the machines and their capabilities.
-
11:09 - 11:12What happens when we couple this machine athleticism
-
11:12 - 11:14with that of a human being?
-
11:14 - 11:17What I have in front of me is a commercial gesture sensor
-
11:17 - 11:19mainly used in gaming.
-
11:19 - 11:20It can recognize what my various body parts
-
11:20 - 11:23are doing in real time.
-
11:23 - 11:25Similar to the pointer that I used earlier,
-
11:25 - 11:27we can use this as inputs to the system.
-
11:27 - 11:30We now have a natural way of interacting
-
11:30 - 11:35with the raw athleticism of these quads with my gestures.
-
12:10 - 12:15(Applause)
-
12:24 - 12:28Interaction doesn't have to be virtual. It can be physical.
-
12:28 - 12:30Take this quad, for example.
-
12:30 - 12:32It's trying to stay at a fixed point in space.
-
12:32 - 12:36If I try to move it out of the way, it fights me,
-
12:36 - 12:40and moves back to where it wants to be.
-
12:40 - 12:43We can change this behavior, however.
-
12:43 - 12:45We can use mathematical models
-
12:45 - 12:48to estimate the force that I'm applying to the quad.
-
12:48 - 12:51Once we know this force, we can also change the laws of physics,
-
12:51 - 12:56as far as the quad is concerned, of course.
-
12:56 - 12:58Here the quad is behaving as if it were
-
12:58 - 13:03in a viscous fluid.
-
13:03 - 13:05We now have an intimate way
-
13:05 - 13:07of interacting with a machine.
-
13:07 - 13:09I will use this new capability to position
-
13:09 - 13:12this camera-carrying quad to the appropriate location
-
13:12 - 13:15for filming the remainder of this demonstration.
-
13:24 - 13:27So we can physically interact with these quads
-
13:27 - 13:29and we can change the laws of physics.
-
13:29 - 13:32Let's have a little bit of fun with this.
-
13:32 - 13:33For what you will see next, these quads
-
13:33 - 13:37will initially behave as if they were on Pluto.
-
13:37 - 13:39As time goes on, gravity will be increased
-
13:39 - 13:41until we're all back on planet Earth,
-
13:41 - 13:43but I assure you we won't get there.
-
13:43 - 13:47Okay, here goes.
-
13:54 - 13:57(Laughter)
-
14:23 - 14:26(Laughter)
-
14:26 - 14:29(Applause)
-
14:29 - 14:31Whew!
-
14:35 - 14:36You're all thinking now,
-
14:36 - 14:38these guys are having way too much fun,
-
14:38 - 14:40and you're probably also asking yourself,
-
14:40 - 14:44why exactly are they building machine athletes?
-
14:44 - 14:47Some conjecture that the role of play in the animal kingdom
-
14:47 - 14:50is to hone skills and develop capabilities.
-
14:50 - 14:52Others think that it has more of a social role,
-
14:52 - 14:53that it's used to bind the group.
-
14:53 - 14:57Similarly, we use the analogy of sports and athleticism
-
14:57 - 14:59to create new algorithms for machines
-
14:59 - 15:01to push them to their limits.
-
15:01 - 15:05What impact will the speed of machines have on our way of life?
-
15:05 - 15:07Like all our past creations and innovations,
-
15:07 - 15:10they may be used to improve the human condition
-
15:10 - 15:13or they may be misused and abused.
-
15:13 - 15:15This is not a technical choice we are faced with;
-
15:15 - 15:16it's a social one.
-
15:16 - 15:18Let's make the right choice,
-
15:18 - 15:20the choice that brings out the best in the future of machines,
-
15:20 - 15:22just like athleticism in sports
-
15:22 - 15:24can bring out the best in us.
-
15:24 - 15:27Let me introduce you to the wizards behind the green curtain.
-
15:27 - 15:30They're the current members of the Flying Machine Arena research team.
-
15:30 - 15:35(Applause)
-
15:35 - 15:38Federico Augugliaro, Dario Brescianini, Markus Hehn,
-
15:38 - 15:41Sergei Lupashin, Mark Muller and Robin Ritz.
-
15:41 - 15:43Look out for them. They're destined for great things.
-
15:43 - 15:44Thank you.
-
15:44 - 15:50(Applause)
- Title:
- The astounding athletic power of quadcopters
- Speaker:
- Raffaello D'Andrea
- Description:
-
In a robot lab at TEDGlobal, Raffaelo D'Andrea demos his flying quadcopters: robots that think like athletes, solving physical problems with algorithms that help them learn. In a series of nifty demos, D'Andrea show drones that play catch, balance and make decisions together -- and watch out for an I-want-this-now demo of Kinect-controlled quads.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 16:08
Camille Martínez edited English subtitles for The astounding athletic power of quadcopters | ||
Krystian Aparta edited English subtitles for The astounding athletic power of quadcopters | ||
Krystian Aparta edited English subtitles for The astounding athletic power of quadcopters | ||
Krystian Aparta commented on English subtitles for The astounding athletic power of quadcopters | ||
Krystian Aparta edited English subtitles for The astounding athletic power of quadcopters | ||
Krystian Aparta edited English subtitles for The astounding athletic power of quadcopters | ||
Morton Bast edited English subtitles for The astounding athletic power of quadcopters | ||
Thu-Huong Ha edited English subtitles for The astounding athletic power of quadcopters |
Krystian Aparta
The English transcript was updated on 2/24/2015.