0:00:11.307,0:00:14.491
So what does it mean for a machine to be athletic?

0:00:14.491,0:00:17.908
We will demonstrate the concept of machine athleticism

0:00:17.908,0:00:19.628
and the research to achieve it

0:00:19.628,0:00:22.139
with the help of these flying machines called quadrocopters,

0:00:22.139,0:00:24.026
or quads, for short.

0:00:26.034,0:00:28.563
Quads have been around for a long time.

0:00:28.563,0:00:30.162
The reason that they're so popular these days

0:00:30.162,0:00:32.127
is because they're mechanically simple.

0:00:32.127,0:00:34.235
By controlling the speeds of these four propellers,

0:00:34.235,0:00:37.283
these machines can roll, pitch, yaw,

0:00:37.283,0:00:40.108
and accelerate along their common orientation.

0:00:40.108,0:00:43.179
On board are also a battery, a computer,

0:00:43.179,0:00:47.133
various sensors and wireless radios.

0:00:47.133,0:00:51.762
Quads are extremely agile, but this agility comes at a cost.

0:00:51.762,0:00:54.754
They are inherently unstable, and they need some form

0:00:54.754,0:00:58.726
of automatic feedback control in order to be able to fly.

0:01:04.259,0:01:07.497
So, how did it just do that?

0:01:07.497,0:01:09.315
Cameras on the ceiling and a laptop

0:01:09.315,0:01:11.891
serve as an indoor global positioning system.

0:01:11.891,0:01:14.076
It's used to locate objects in the space

0:01:14.076,0:01:16.603
that have these reflective markers on them.

0:01:16.603,0:01:18.664
This data is then sent to another laptop

0:01:18.664,0:01:20.875
that is running estimation and control algorithms,

0:01:20.875,0:01:22.761
which in turn sends commands to the quad,

0:01:22.761,0:01:26.246
which is also running estimation and control algorithms.

0:01:29.620,0:01:32.130
The bulk of our research is algorithms.

0:01:32.130,0:01:36.291
It's the magic that brings these machines to life.

0:01:36.291,0:01:38.334
So how does one design the algorithms

0:01:38.334,0:01:40.611
that create a machine athlete?

0:01:40.611,0:01:43.491
We use something broadly called model-based design.

0:01:43.491,0:01:46.645
We first capture the physics with a mathematical model

0:01:46.645,0:01:48.779
of how the machines behave.

0:01:48.779,0:01:51.076
We then use a branch of mathematics

0:01:51.076,0:01:53.907
called control theory to analyze these models

0:01:53.907,0:01:57.820
and also to synthesize algorithms for controlling them.

0:01:57.820,0:02:00.868
For example, that's how we can make the quad hover.

0:02:00.868,0:02:02.276
We first captured the dynamics

0:02:02.276,0:02:04.182
with a set of differential equations.

0:02:04.182,0:02:06.531
We then manipulate these equations with the help

0:02:06.531,0:02:11.171
of control theory to create algorithms that stabilize the quad.

0:02:11.171,0:02:14.117
Let me demonstrate the strength of this approach.

0:02:17.198,0:02:19.747
Suppose that we want this quad to not only hover

0:02:19.747,0:02:22.529
but to also balance this pole.

0:02:22.529,0:02:24.373
With a little bit of practice,

0:02:24.373,0:02:26.764
it's pretty straightforward for a human being to do this,

0:02:26.764,0:02:28.519
although we do have the advantage of having

0:02:28.519,0:02:29.847
two feet on the ground

0:02:29.847,0:02:32.804
and the use of our very versatile hands.

0:02:32.804,0:02:35.273
It becomes a little bit more difficult

0:02:35.273,0:02:37.569
when I only have one foot on the ground

0:02:37.569,0:02:40.368
and when I don't use my hands.

0:02:40.368,0:02:43.319
Notice how this pole has a reflective marker on top,

0:02:43.319,0:02:47.213
which means that it can be located in the space.

0:02:52.971,0:02:58.947
(Applause)

0:02:58.947,0:03:01.773
You can notice that this quad is making fine adjustments

0:03:01.773,0:03:03.968
to keep the pole balanced.

0:03:03.968,0:03:07.027
How did we design the algorithms to do this?

0:03:07.027,0:03:09.132
We added the mathematical model of the pole

0:03:09.132,0:03:10.651
to that of the quad.

0:03:10.651,0:03:13.605
Once we have a model of the combined quad-pole system,

0:03:13.605,0:03:18.538
we can use control theory to create algorithms for controlling it.

0:03:18.538,0:03:20.126
Here, you see that it's stable,

0:03:20.126,0:03:22.819
and even if I give it little nudges,

0:03:22.819,0:03:28.222
it goes back to the nice, balanced position.

0:03:28.222,0:03:29.885
We can also augment the model to include

0:03:29.885,0:03:32.326
where we want the quad to be in space.

0:03:32.326,0:03:35.395
Using this pointer, made out of reflective markers,

0:03:35.395,0:03:37.619
I can point to where I want the quad to be in space

0:03:37.619,0:03:40.986
a fixed distance away from me.

0:03:55.763,0:03:58.665
The key to these acrobatic maneuvers is algorithms,

0:03:58.665,0:04:01.091
designed with the help of mathematical models

0:04:01.091,0:04:03.041
and control theory.

0:04:03.041,0:04:05.255
Let's tell the quad to come back here

0:04:05.255,0:04:07.074
and let the pole drop,

0:04:07.074,0:04:08.992
and I will next demonstrate the importance

0:04:08.992,0:04:11.111
of understanding physical models

0:04:11.111,0:04:15.231
and the workings of the physical world.

0:04:25.462,0:04:27.232
Notice how the quad lost altitude

0:04:27.232,0:04:29.314
when I put this glass of water on it.

0:04:29.314,0:04:32.115
Unlike the balancing pole, I did not include

0:04:32.115,0:04:34.699
the mathematical model of the glass in the system.

0:04:34.699,0:04:37.618
In fact, the system doesn't even know that the glass of water is there.

0:04:37.618,0:04:40.910
Like before, I could use the pointer to tell the quad

0:04:40.910,0:04:43.499
where I want it to be in space.

0:04:43.499,0:04:53.096
(Applause)

0:04:53.096,0:04:55.206
Okay, you should be asking yourself,

0:04:55.206,0:04:57.715
why doesn't the water fall out of the glass?

0:04:57.715,0:05:00.926
Two facts: The first is that gravity acts

0:05:00.926,0:05:03.051
on all objects in the same way.

0:05:03.051,0:05:05.515
The second is that the propellers are all pointing

0:05:05.515,0:05:08.642
in the same direction of the glass, pointing up.

0:05:08.642,0:05:11.019
You put these two things together, the net result

0:05:11.019,0:05:13.306
is that all side forces on the glass are small

0:05:13.306,0:05:16.091
and are mainly dominated by aerodynamic effects,

0:05:16.091,0:05:19.932
which as these speeds are negligible.

0:05:23.336,0:05:25.407
And that's why you don't need to model the glass.

0:05:25.407,0:05:29.255
It naturally doesn't spill no matter what the quad does.

0:05:38.651,0:05:45.702
(Applause)

0:05:45.702,0:05:49.718
The lesson here is that some high-performance tasks

0:05:49.718,0:05:51.115
are easier than others,

0:05:51.115,0:05:53.411
and that understanding the physics of the problem

0:05:53.411,0:05:56.027
tells you which ones are easy and which ones are hard.

0:05:56.027,0:05:58.386
In this instance, carrying a glass of water is easy.

0:05:58.386,0:06:02.171
Balancing a pole is hard.

0:06:02.171,0:06:03.973
We've all heard stories of athletes

0:06:03.973,0:06:06.466
performing feats while physically injured.

0:06:06.466,0:06:07.985
Can a machine also perform

0:06:07.985,0:06:10.636
with extreme physical damage?

0:06:10.636,0:06:12.362
Conventional wisdom says that you need

0:06:12.362,0:06:16.052
at least four fixed motor propeller pairs in order to fly,

0:06:16.052,0:06:18.164
because there are four degrees of freedom to control:

0:06:18.164,0:06:21.192
roll, pitch, yaw and acceleration.

0:06:21.192,0:06:24.437
Hexacopters and octocopters, with six and eight propellers,

0:06:24.437,0:06:25.920
can provide redundancy,

0:06:25.920,0:06:27.909
but quadrocopters are much more popular

0:06:27.909,0:06:29.724
because they have the minimum number

0:06:29.724,0:06:32.200
of fixed motor propeller pairs: four.

0:06:32.200,0:06:34.165
Or do they?

0:06:49.476,0:06:52.076
If we analyze the mathematical model of this machine

0:06:52.076,0:06:54.252
with only two working propellers,

0:06:54.252,0:07:01.008
we discover that there's an unconventional way to fly it.

0:07:07.980,0:07:09.686
We relinquish control of yaw,

0:07:09.686,0:07:12.756
but roll, pitch and acceleration can still be controlled

0:07:12.756,0:07:18.208
with algorithms that exploit this new configuration.

0:07:21.668,0:07:24.100
Mathematical models tell us exactly when

0:07:24.100,0:07:26.172
and why this is possible.

0:07:26.172,0:07:28.540
In this instance, this knowledge allows us to design

0:07:28.540,0:07:30.796
novel machine architectures

0:07:30.796,0:07:34.972
or to design clever algorithms that gracefully handle damage,

0:07:34.972,0:07:36.696
just like human athletes do,

0:07:36.696,0:07:40.519
instead of building machines with redundancy.

0:07:40.519,0:07:42.631
We can't help but hold our breath when we watch

0:07:42.631,0:07:45.068
a diver somersaulting into the water,

0:07:45.068,0:07:46.858
or when a vaulter is twisting in the air,

0:07:46.858,0:07:48.576
the ground fast approaching.

0:07:48.576,0:07:51.144
Will the diver be able to pull off a rip entry?

0:07:51.144,0:07:53.144
Will the vaulter stick the landing?

0:07:53.144,0:07:54.812
Suppose we want this quad here

0:07:54.812,0:07:57.263
to perform a triple flip and finish off

0:07:57.263,0:07:59.532
at the exact same spot that it started.

0:07:59.532,0:08:01.879
This maneuver is going to happen so quickly

0:08:01.879,0:08:05.508
that we can't use position feedback to correct the motion during execution.

0:08:05.508,0:08:07.782
There simply isn't enough time.

0:08:07.782,0:08:11.476
Instead, what the quad can do is perform the maneuver blindly,

0:08:11.476,0:08:13.708
observe how it finishes the maneuver,

0:08:13.708,0:08:16.132
and then use that information to modify its behavior

0:08:16.132,0:08:18.421
so that the next flip is better.

0:08:18.421,0:08:20.233
Similar to the diver and the vaulter,

0:08:20.233,0:08:22.152
it is only through repeated practice

0:08:22.152,0:08:24.077
that the maneuver can be learned and executed

0:08:24.077,0:08:26.210
to the highest standard.

0:08:34.412,0:08:39.164
(Applause)

0:08:39.164,0:08:42.636
Striking a moving ball is a necessary skill in many sports.

0:08:42.636,0:08:44.431
How do we make a machine do

0:08:44.431,0:08:48.184
what an athlete does seemingly without effort?

0:09:03.943,0:09:10.542
(Applause)

0:09:10.542,0:09:13.192
This quad has a racket strapped onto its head

0:09:13.192,0:09:16.679
with a sweet spot roughly the size of an apple, so not too large.

0:09:16.679,0:09:19.846
The following calculations are made every 20 milliseconds,

0:09:19.846,0:09:21.602
or 50 times per second.

0:09:21.602,0:09:24.414
We first figure out where the ball is going.

0:09:24.414,0:09:26.851
We then next calculate how the quad should hit the ball

0:09:26.851,0:09:29.950
so that it flies to where it was thrown from.

0:09:29.950,0:09:34.406
Third, a trajectory is planned that carries the quad

0:09:34.406,0:09:37.430
from its current state to the impact point with the ball.

0:09:37.430,0:09:41.239
Fourth, we only execute 20 milliseconds' worth of that strategy.

0:09:41.239,0:09:44.046
Twenty milliseconds later, the whole process is repeated

0:09:44.046,0:09:46.289
until the quad strikes the ball.

0:09:55.560,0:09:58.206
(Applause)

0:09:58.206,0:10:01.578
Machines can not only perform dynamic maneuvers on their own,

0:10:01.578,0:10:03.441
they can do it collectively.

0:10:03.441,0:10:07.028
These three quads are cooperatively carrying a sky net.

0:10:16.574,0:10:21.985
(Applause)

0:10:21.985,0:10:24.182
They perform an extremely dynamic

0:10:24.182,0:10:26.350
and collective maneuver

0:10:26.350,0:10:28.042
to launch the ball back to me.

0:10:28.042,0:10:31.553
Notice that, at full extension, these quads are vertical.

0:10:36.486,0:10:38.258
(Applause)

0:10:38.258,0:10:40.545
In fact, when fully extended,

0:10:40.545,0:10:43.230
this is roughly five times greater than what a bungee jumper feels

0:10:43.230,0:10:47.818
at the end of their launch.

0:10:51.009,0:10:53.686
The algorithms to do this are very similar

0:10:53.686,0:10:57.070
to what the single quad used to hit the ball back to me.

0:10:57.070,0:10:59.934
Mathematical models are used to continuously re-plan

0:10:59.934,0:11:04.449
a cooperative strategy 50 times per second.

0:11:04.449,0:11:06.326
Everything we have seen so far has been

0:11:06.326,0:11:08.789
about the machines and their capabilities.

0:11:08.789,0:11:11.804
What happens when we couple this machine athleticism

0:11:11.804,0:11:13.580
with that of a human being?

0:11:13.580,0:11:17.385
What I have in front of me is a commercial gesture sensor

0:11:17.385,0:11:18.694
mainly used in gaming.

0:11:18.694,0:11:20.498
It can recognize what my various body parts

0:11:20.498,0:11:22.546
are doing in real time.

0:11:22.546,0:11:24.662
Similar to the pointer that I used earlier,

0:11:24.662,0:11:27.454
we can use this as inputs to the system.

0:11:27.454,0:11:29.850
We now have a natural way of interacting

0:11:29.850,0:11:34.794
with the raw athleticism of these quads with my gestures.

0:12:10.453,0:12:14.853
(Applause)

0:12:23.982,0:12:27.989
Interaction doesn't have to be virtual. It can be physical.

0:12:27.989,0:12:29.714
Take this quad, for example.

0:12:29.714,0:12:32.342
It's trying to stay at a fixed point in space.

0:12:32.342,0:12:36.179
If I try to move it out of the way, it fights me,

0:12:36.179,0:12:40.334
and moves back to where it wants to be.

0:12:40.334,0:12:43.373
We can change this behavior, however.

0:12:43.373,0:12:45.072
We can use mathematical models

0:12:45.072,0:12:48.205
to estimate the force that I'm applying to the quad.

0:12:48.205,0:12:51.443
Once we know this force, we can also change the laws of physics,

0:12:51.443,0:12:55.942
as far as the quad is concerned, of course.

0:12:55.942,0:12:58.205
Here the quad is behaving as if it were

0:12:58.205,0:13:02.570
in a viscous fluid.

0:13:02.570,0:13:04.744
We now have an intimate way

0:13:04.744,0:13:06.593
of interacting with a machine.

0:13:06.593,0:13:09.099
I will use this new capability to position

0:13:09.099,0:13:11.676
this camera-carrying quad to the appropriate location

0:13:11.676,0:13:14.862
for filming the remainder of this demonstration.

0:13:24.222,0:13:26.990
So we can physically interact with these quads

0:13:26.990,0:13:29.393
and we can change the laws of physics.

0:13:29.393,0:13:31.706
Let's have a little bit of fun with this.

0:13:31.706,0:13:33.410
For what you will see next, these quads

0:13:33.410,0:13:36.906
will initially behave as if they were on Pluto.

0:13:36.906,0:13:39.218
As time goes on, gravity will be increased

0:13:39.218,0:13:41.338
until we're all back on planet Earth,

0:13:41.338,0:13:43.057
but I assure you we won't get there.

0:13:43.057,0:13:46.558
Okay, here goes.

0:13:53.667,0:13:57.079
(Laughter)

0:14:22.778,0:14:25.986
(Laughter)

0:14:25.986,0:14:29.372
(Applause)

0:14:29.372,0:14:30.579
Whew!

0:14:34.779,0:14:36.296
You're all thinking now,

0:14:36.296,0:14:38.271
these guys are having way too much fun,

0:14:38.271,0:14:40.350
and you're probably also asking yourself,

0:14:40.350,0:14:44.286
why exactly are they building machine athletes?

0:14:44.286,0:14:46.938
Some conjecture that the role of play in the animal kingdom

0:14:46.938,0:14:49.690
is to hone skills and develop capabilities.

0:14:49.690,0:14:51.756
Others think that it has more of a social role,

0:14:51.756,0:14:53.430
that it's used to bind the group.

0:14:53.430,0:14:56.963
Similarly, we use the analogy of sports and athleticism

0:14:56.963,0:14:59.042
to create new algorithms for machines

0:14:59.042,0:15:01.189
to push them to their limits.

0:15:01.189,0:15:04.866
What impact will the speed of machines have on our way of life?

0:15:04.866,0:15:07.255
Like all our past creations and innovations,

0:15:07.255,0:15:10.082
they may be used to improve the human condition

0:15:10.082,0:15:12.583
or they may be misused and abused.

0:15:12.583,0:15:14.506
This is not a technical choice we are faced with;

0:15:14.506,0:15:16.261
it's a social one.

0:15:16.261,0:15:17.651
Let's make the right choice,

0:15:17.651,0:15:20.056
the choice that brings out the best in the future of machines,

0:15:20.056,0:15:21.868
just like athleticism in sports

0:15:21.868,0:15:24.029
can bring out the best in us.

0:15:24.029,0:15:27.479
Let me introduce you to the wizards behind the green curtain.

0:15:27.479,0:15:30.379
They're the current members of the Flying Machine Arena research team.

0:15:30.379,0:15:35.148
(Applause)

0:15:35.148,0:15:38.194
Federico Augugliaro, Dario Brescianini, Markus Hehn,

0:15:38.194,0:15:41.018
Sergei Lupashin, Mark Muller and Robin Ritz.

0:15:41.018,0:15:42.950
Look out for them. They're destined for great things.

0:15:42.950,0:15:44.128
Thank you.

0:15:44.128,0:15:50.482
(Applause)