Robots that fly ... and cooperate | Vijay Kumar

翻译人员: Xiaoqiao Xie 校对人员: Angelia King

Good morning.

早上好

I'm here today to talk about autonomous flying beach balls.

我今天想谈谈

自主飞行沙滩球

(Laughter)

其实，是小型飞行器，像这一个

No, agile aerial robots like this one.

I'd like to tell you a little bit about the challenges in building these,

我想和大家谈谈设计这些飞行器时的挑战

和使用这些飞行器能给我们带来的

and some of the terrific opportunities for applying this technology.

很多用处

So these robots are related to unmanned aerial vehicles.

这些飞行器

源于无人驾驶的飞行器

但是那些都体积很大

However, the vehicles you see here are big.

通常上万磅重

They weigh thousands of pounds, are not by any means agile.

毫无灵活型可言

They're not even autonomous.

它们也不是真的自主飞行的

In fact, many of these vehicles are operated by flight crews

事实上，很多这些飞行器

都是受飞行团队控制的

包括好几个飞行员

that can include multiple pilots,

operators of sensors,

感应雷达操作员

and mission coordinators.

和团队协调员

What we're interested in is developing robots like this --

我们想设计的飞行器是这样的——

这里有两张照片——

and here are two other pictures --

是你能够在超市里买到的那种小飞行器

of robots that you can buy off the shelf.

小型直升机，四个螺旋桨

So these are helicopters with four rotors,

不超过一米长

and they're roughly a meter or so in scale,

只不过几磅重

and weigh several pounds.

我们把它们稍微改造一下，加上感应器和处理器，

And so we retrofit these with sensors and processors,

它们就可以在室内飞

and these robots can fly indoors.

用不着导航系统

Without GPS.

我现在拿着的这个飞行器

The robot I'm holding in my hand

是其中之一

is this one,

是两个学生做出来的

and it's been created by two students,

艾利克斯和丹尼尔

Alex and Daniel.

这个仅仅比零点一磅

So this weighs a little more than a tenth of a pound.

稍微重一点

It consumes about 15 watts of power.

只需要大约十五瓦的电源

你能看到

And as you can see, it's about eight inches in diameter.

它的直径大约只有八个英寸

让我给你们快速解释一下

So let me give you just a very quick tutorial

on how these robots work.

这些飞行器是怎么工作的

它有四个螺旋桨

So it has four rotors.

If you spin these rotors at the same speed,

当四个螺旋桨转速相同

the robot hovers.

这个飞行器就浮在空中

If you increase the speed of each of these rotors,

当所有螺旋桨的速度提升时

这个飞行器就加速升高

then the robot flies up, it accelerates up.

Of course, if the robot were tilted,

当然了，如果飞行器已经是倾斜的

向着地平线侧过来

inclined to the horizontal,

then it would accelerate in this direction.

就会向这个方向加速

So to get it to tilt,

怎么能让它侧过来呢，有两个途径

there's one of two ways of doing it.

从这张照片

So in this picture, you see that rotor four is spinning faster

你能看到四号螺旋桨旋转加速

and rotor two is spinning slower.

同时二号螺旋桨转速变慢

And when that happens,

这时

there's a moment that causes this robot to roll.

飞行器就能向一边倒

反之亦然

And the other way around,

if you increase the speed of rotor three and decrease the speed of rotor one,

当三号螺旋桨加速

一号减速时

飞行器就向前倒

then the robot pitches forward.

And then finally,

最后

if you spin opposite pairs of rotors

如果任意两端的螺旋桨的转速

faster than the other pair,

大于另两端的螺旋桨的转速

then the robot yaws about the vertical axis.

飞行器就能原地旋转

所以装在飞行器上的处理器

So an on-board processor

essentially looks at what motions need to be executed

基本上能判断需要执行哪些动作

然后把它们组合起来

and combines these motions,

决定给螺旋桨下什么指令

and figures out what commands to send to the motors --

一秒钟六百次

600 times a second.

That's basically how this thing operates.

简单地说这些飞行器就是这么工作的

这个设计的一个好处

So one of the advantages of this design

就是小巧

is when you scale things down,

这些飞行器很灵活

the robot naturally becomes agile.

这里的R

So here, R is the characteristic length of the robot.

是飞行器的长度

其实是半径

It's actually half the diameter.

And there are lots of physical parameters that change as you reduce R.

当半径变小时

很多物理参数都会变

The one that's most important is the inertia,

最重要的一个参数是

惯性， 也就是对于运动的阻力

or the resistance to motion.

结果是

So it turns out the inertia, which governs angular motion,

惯性决定角速度

它是半径的五次方函数

scales as a fifth power of R.

当半径变得越来越小时

So the smaller you make R,

the more dramatically the inertia reduces.

惯性越来越快地减小

So as a result, the angular acceleration,

另一个结果是角速度的加速度

denoted by the Greek letter alpha here,

也就是这里的希腊字母alpha

goes as 1 over R.

等于一除以半径

It's inversely proportional to R.

也就是半径的倒数

The smaller you make it, the more quickly you can turn.

当半径越小时飞行器能转弯越快

这个视频清楚地显示

So this should be clear in these videos.

大家看右下角的飞行器

On the bottom right, you see a robot performing a 360-degree flip

正在做一个三百六十度翻转

in less than half a second.

只需要不到半秒

Multiple flips, a little more time.

连续翻转，稍微时间长一点

这里飞行器上用的处理器

So here the processes on board

能够从飞行器上的加速度计

are getting feedback from accelerometers and gyros on board,

和陀螺仪得到反馈信息

and calculating, like I said before,

然后算出，就像我刚才讲的

commands at 600 times a second,

一秒钟六百个指令

to stabilize this robot.

来稳定控制这个飞行器

So on the left, you see Daniel throwing this robot up into the air,

在左边你能看到丹尼尔把飞行器抛到空中

and it shows you how robust the control is.

你能看到飞行器的控制有多快

不管你怎么扔

No matter how you throw it,

the robot recovers and comes back to him.

飞行器都能恢复平衡飞回来

So why build robots like this?

为什么我们要设计这种飞行器呢？

因为这样的飞行器有很多用处

Well, robots like this have many applications.

你能把它们放进像这样的大楼里

You can send them inside buildings like this,

as first responders to look for intruders,

作为报警器去寻找入侵者

寻找生化泄漏

maybe look for biochemical leaks,

或者煤气泄漏

gaseous leaks.

你还能用它们

You can also use them for applications like construction.

建摩天大楼呢

So here are robots carrying beams, columns

这里是飞行器在搬梁运柱

架构一个立方体的建筑

and assembling cube-like structures.

I'll tell you a little bit more about this.

这里我想和大家介绍一下

The robots can be used for transporting cargo.

这些机器人能被用来运货

So one of the problems with these small robots

当然一个问题是这些小飞行器

is their payload-carrying capacity.

担不了多少重量

So you might want to have multiple robots carry payloads.

你可能需要很多飞行器

来搬运重物

This is a picture of a recent experiment we did --

我们新做了个实验——

其实不那么新了——

actually not so recent anymore --

in Sendai, shortly after the earthquake.

在日本仙台，地震后不久

So robots like this could be sent into collapsed buildings,

我们能把这些飞行器

送进倒塌的楼房

to assess the damage after natural disasters,

或者核反应堆大楼

or sent into reactor buildings,

to map radiation levels.

来探测放射性强度

So one fundamental problem that the robots have to solve

一个根本的问题

是当这些飞行器需要自控飞行，

if they are to be autonomous,

is essentially figuring out how to get from point A to point B.

它们自己得弄明白

怎么从一个地点到另一个地点

So this gets a little challenging,

这就变得有点难度了

because the dynamics of this robot are quite complicated.

因为这些飞行器的动力学是很复杂的

In fact, they live in a 12-dimensional space.

事实上它们总在对付十二维的空间

这里我们用了一点小技巧

So we use a little trick.

We take this curved 12-dimensional space,

我们拿这个十二位的空间

把它们转换成

and transform it into a flat, four-dimensional space.

平的四维空间

这个四维空间

And that four-dimensional space consists of X, Y, Z,

包括了横轴，纵轴和竖轴，还有旋转轴

and then the yaw angle.

And so what the robot does,

这些飞行器只需要

is it plans what we call a minimum-snap trajectory.

计划一件事，我们管它叫最小化加加加速度轨道

提醒大家一点点物理学

So to remind you of physics:

You have position, derivative, velocity;

这里我们有位置向量，导数，速度

then acceleration;

和加速度

and then comes jerk,

还有加加速度

and then comes snap.

还有加加加速度

So this robot minimizes snap.

这个飞行器把加加加速度最小化

So what that effectively does,

基本上它的工作是

is produce a smooth and graceful motion.

创造一个光滑优雅的运动曲线

And it does that avoiding obstacles.

这样来绕开障碍物

所以这个四维平面中，这个飞行器使用

So these minimum-snap trajectories in this flat space are then transformed

最小化加加加速度轨道， 然后转换回到

back into this complicated 12-dimensional space,

复杂的十二维空间

飞行器必须这样做来

which the robot must do for control and then execution.

获得控制和执行动作

So let me show you some examples

让我给大家看几个例子

of what these minimum-snap trajectories look like.

这些最小化加加加速度轨道是什么样的

这是第一个视频

And in the first video,

you'll see the robot going from point A to point B,

这个飞行器从一个地点飞到另一个地点

中间经停一下

through an intermediate point.

(Whirring noise)

显然这个飞行器能

So the robot is obviously capable of executing any curve trajectory.

飞出一个曲线轨道

还有这样的打圈的轨道

So these are circular trajectories,

where the robot pulls about two G's.

这里飞行器对抗两倍的重力

Here you have overhead motion capture cameras on the top

它们上方还有一个动感监控摄像机，每秒一百幅画面

that tell the robot where it is 100 times a second.

来告诉这些飞行器它们的位置

It also tells the robot where these obstacles are.

也能告诉这些飞行器障碍物在哪里

障碍物移动都不要紧

And the obstacles can be moving.

And here, you'll see Daniel throw this hoop into the air,

当丹尼尔把套圈扔到空中

while the robot is calculating the position of the hoop,

飞行器就开始计算套圈的位置

试图预测怎么才能最有效地钻过去

and trying to figure out how to best go through the hoop.

作为一个科研人员

So as an academic,

we're always trained to be able to jump through hoops

我们总在试图钻出重重圈套，拿到更多经费

to raise funding for our labs,

甚至训练了我们的飞行器也来做这个

and we get our robots to do that.

(Applause)

（掌声）

另一个飞行器能做的事情

So another thing the robot can do

是当我们预先编入一些轨迹

is it remembers pieces of trajectory

that it learns or is pre-programmed.

或者它自己学着走过的，它能够记住

这里大家能看到

So here, you see the robot combining a motion that builds up momentum,

飞行器能够（在预设轨迹上）加上一个动作

积聚动量

and then changes its orientation and then recovers.

改变它的定向，再回到预设轨迹上来

它必须这样做因为这个窗上的缝隙

So it has to do this because this gap in the window

is only slightly larger than the width of the robot.

只比它的宽度大一点点

所以就像是一个跳水运动员

So just like a diver stands on a springboard

and then jumps off it to gain momentum,

从跳板上起跳，聚集动量，

做个旋转，两圈半

and then does this pirouette, this two and a half somersault through

然后优雅地回到平衡

and then gracefully recovers,

this robot is basically doing that.

这个飞行器是自主这样做的

So it knows how to combine little bits and pieces of trajectories

它知道怎么把小段的轨迹组合起来

to do these fairly difficult tasks.

来做这些高难度的技巧

现在我想换个话题谈谈这些小型飞行器

So I want change gears.

So one of the disadvantages of these small robots is its size.

的不足之处，就是体积小

我已经提过

And I told you earlier

that we may want to employ lots and lots of robots

我们需要使用很多飞行器

来克服体积小的不便

to overcome the limitations of size.

一个难点是

So one difficulty is:

How do you coordinate lots of these robots?

怎么使得这些飞行器集体飞行？

And so here, we looked to nature.

我们在大自然中寻找答案

So I want to show you a clip of Aphaenogaster desert ants,

我想给大家看一个视频

是关于Aphaenogaster沙漠蚁的

在史狄文·普热特教授的实验室里，这些蚂蚁一起搬运重物

in Professor Stephen Pratt's lab, carrying an object.

这是一个无花果

So this is actually a piece of fig.

事实上无论什么东西，只要蘸上无花果汁

Actually you take any object coated with fig juice,

这些蚂蚁都会把它们带回巢去

and the ants will carry it back to the nest.

So these ants don't have any central coordinator.

这些蚂蚁没有任何中央调控

它们是靠感应邻近的蚂蚁

They sense their neighbors.

它们也没有明确的交流

There's no explicit communication.

但是因为它们能够感应邻近的蚂蚁

But because they sense the neighbors

也能感应抬着的重物

and because they sense the object,

they have implicit coordination across the group.

整群的蚂蚁有默契

这样的协调

So this is the kind of coordination we want our robots to have.

正是飞行器需要的

So when we have a robot which is surrounded by neighbors --

当一个飞行器

被其他飞行器环绕时——

让我们注意 I 和 J 这两个——

and let's look at robot I and robot J --

当它们成群飞行时

what we want the robots to do,

我们希望这两个飞行器

is to monitor the separation between them,

as they fly in formation.

能够监控它们之间的距离

And then you want to make sure

我们需要确定

that this separation is within acceptable levels.

这个距离是在可接受的范围里的

飞行器要检测这个变化

So again, the robots monitor this error

and calculate the control commands 100 times a second,

在控制指令中计算进去

也是每秒一百次

which then translates into motor commands,

这个控制指令每秒会被送到马达六百次

600 times a second.

所以这个程序

So this also has to be done in a decentralized way.

是分散化执行的

Again, if you have lots and lots of robots,

再有，如果你有很多很多飞行器

要完成集体飞行任务，能足够快地集中协调所有这些信息

it's impossible to coordinate all this information centrally

fast enough in order for the robots to accomplish the task.

是几乎不可能的

Plus, the robots have to base their actions only on local information --

加上这些飞行器只能

依靠局部的信息来决定做什么动作

也就是要靠感应邻近的飞行器

what they sense from their neighbors.

最后

And then finally,

we insist that the robots be agnostic to who their neighbors are.

我们希望这些机器人

不知道它们的邻居是谁

So this is what we call anonymity.

也就是匿名飞行

下一个我想给大家展示的

So what I want to show you next is a video of 20 of these little robots,

是这段视频

这二十个小型飞行器

flying in formation.

成群飞行

它们在监测邻居的位置

They're monitoring their neighbors' positions.

维持群队

They're maintaining formation.

The formations can change.

群队的形状还能变

They can be planar formations,

它们可以在一个平面上飞

they can be three-dimensional formations.

也可以上中下地飞

大家可以看到

As you can see here,

they collapse from a three-dimensional formation into planar formation.

它们能从上中下的群队变成平面的

在飞越障碍物的时候

And to fly through obstacles,

they can adapt the formations on the fly.

它们能边飞边变换队形

我想强调，这些飞行器距离都很近

So again, these robots come really close together.

As you can see in this figure-eight flight,

比如这个群队，八架飞行器

they come within inches of each other.

相互距离不过几英寸

尽管在空气动力学上

And despite the aerodynamic interactions with these propeller blades,

这些螺旋桨相互干扰

they're able to maintain stable flight.

它们还是能够维持平稳飞行

(Applause)

（掌声）

现在它们会成群飞了

So once you know how to fly in formation,

它们就可以合作抬重物

you can actually pick up objects cooperatively.

这里展示的是

So this just shows that we can double, triple, quadruple

我们能够把飞行器的能力

翻倍，翻三倍，四倍

the robots' strength,

by just getting them to team with neighbors, as you can see here.

仅仅通过让它们和邻居合作，大家可以看到

这样做的一个不便之处

One of the disadvantages of doing that is, as you scale things up --

就是当加大数量时——

so if you have lots of robots carrying the same thing,

比如使用很多飞行器来抬一个物体

你其实是加大了惯性

you're essentially increasing the inertia,

and therefore you pay a price; they're not as agile.

这样它们就不够灵活了，这是一个代价

But you do gain in terms of payload-carrying capacity.

但是你可以增加载荷承载量

另一个我想给大家展示的用处是——

Another application I want to show you -- again, this is in our lab.

这是在我们实验室

This is work done by Quentin Lindsey, who's a graduate student.

这是研究生昆汀·林夕的工作

他的算法程序告诉这些飞行器

So his algorithm essentially tells these robots

怎么使用桁架结构

how to autonomously build cubic structures

自动建造

一个立方体

from truss-like elements.

他的算法程序告诉这些机器人

So his algorithm tells the robot what part to pick up,

该用哪一块

什么时候用，用在哪里

when, and where to place it.

从这个视频我们可以看到——

So in this video you see --

and it's sped up 10, 14 times --

这个视频是十倍或者十四倍速度播放的——

大家可以看到飞行器在搭建很不一样的构架

you see three different structures being built by these robots.

并且，所有的运动都是自主的

And again, everything is autonomous,

昆汀仅仅是

and all Quentin has to do

is to give them a blueprint of the design that he wants to build.

给它们一个蓝图

也就是他想建的设计

So all these experiments you've seen thus far,

所有这里展示的实验

all these demonstrations,

所有这些演习

have been done with the help of motion-capture systems.

都是靠着它们自己的动感检测摄像机完成的

那么，当它们离开实验室

So what happens when you leave your lab,

来到真实世界的时候，又怎么样呢？

and you go outside into the real world?

And what if there's no GPS?

没有卫星导航会怎么样？

So this robot is actually equipped with a camera,

这个飞行器

其实装有一个摄像机

和一个激光测距仪，一个激光扫描仪

and a laser rangefinder, laser scanner.

它可以使用这些探测装置

And it uses these sensors to build a map of the environment.

来描绘周围的环境的地图

这个地图包括很多细节——

What that map consists of are features --

玄关，窗户

like doorways, windows, people, furniture --

人，家具——

还能弄清楚相对于这些东西

and it then figures out where its position is,

它自己在哪里

with respect to the features.

So there is no global coordinate system.

所以这里没有整体的协调系统

这个协调系统是靠飞行器自己来完成的

The coordinate system is defined based on the robot,

where it is and what it's looking at.

它自己在哪里，前面有什么

And it navigates with respect to those features.

还能利用周围环境为自己找到出路

这里我想给大家再看一段视频

So I want to show you a clip

of algorithms developed by Frank Shen and Professor Nathan Michael,

这个算法程序是法兰克·沈

和南希·麦克教授编的

that shows this robot entering a building for the very first time,

当这个飞行器第一次飞入一个建筑

and creating this map on the fly.

它是怎么边飞边画地图的

So the robot then figures out what the features are,

这个飞行器弄明白了这些细节

it builds the map,

开始画地图

it figures out where it is with respect to the features,

弄明白了相对这些细节，自己在哪里，

and then estimates its position 100 times a second,

然后自我定位

全以每秒一百次的速度发生

allowing us to use the control algorithms that I described to you earlier.

这就给我们一个机会来控制这些算法

像我之前讲过的

So this robot is actually being commanded remotely by Frank,

所以这个机器人其实是

被法兰克遥控的

但是它自己也可以弄明白

but the robot can also figure out where to go on its own.

怎么飞

假设我想放一个这样的飞行器进一幢楼

So suppose I were to send this into a building,

我并不知道里面是什么样的

and I had no idea what this building looked like.

我可以让它飞进去

I can ask this robot to go in,

创造一个地图

create a map,

然后飞回来告诉我里面是什么样的

and then come back and tell me what the building looks like.

So here, the robot is not only solving the problem

所以，这个飞行器不仅仅解决了

怎么从一点到另一点的问题

of how to go from point A to point B in this map,

but it's figuring out what the best point B is at every time.

还能够随时知道

最好的目标在哪里

基本上，它知道该去搜索哪里

So essentially it knows where to go

to look for places that have the least information,

因为那里的信息是最“未知”的

这就是它怎么填充这个地图

and that's how it populates this map.

So I want to leave you with one last application.

这里我想展示给大家

最后一个用途

And there are many applications of this technology.

当然这个技术有很多很多用途

I'm a professor, and we're passionate about education.

我是个教授，我们很关心教育

这样的飞行器其实可以改变

Robots like this can really change the way we do K-12 education.

我们的小学和中学教育

我们在南加州

But we're in Southern California,

离洛杉矶很近

close to Los Angeles,

所以我不得不

so I have to conclude with something focused on entertainment.

放点娱乐元素进去

我想给大家看一个音乐视频

I want to conclude with a music video.

我想向你们介绍艾利克斯和丹尼尔，

I want to introduce the creators, Alex and Daniel, who created this video.

他们是导演兼制作

（掌声）

(Applause)

So before I play this video,

在我播放这个视频前

I want to tell you that they created it in the last three days,

我想告诉大家这是他们在过去三天做出来的

after getting a call from Chris.

因为主持人克瑞斯给我打了个电话

And the robots that play in the video are completely autonomous.

在这个视频中表演的飞行器

全是靠自控表演的

You will see nine robots play six different instruments.

你能看到九个机器人，演奏六种不同乐器

当然了，这是为了今年的TED2012特别制作的

And of course, it's made exclusively for TED 2012.

请欣赏

Let's watch.

(Sound of air escaping from valve)

(Music)

(Whirring sound)

(Music)

（音乐）

（掌声）

(Applause) (Cheers)