How do self-driving cars “see”? - Sajan Saini

譯者: Lilian Chiu 審譯者: Helen Chang

It’s late, pitch dark, and a self-driving car winds down a narrow country road.

時間很晚了，天很黑，

有一台自動駕駛汽車 在鄉村道路上迂迴行駛。

Suddenly, three hazards appear at the same time.

突然，三樣危險物同時出現。

What happens next?

接下來會發生什麼事？

Before it can navigate this onslaught of obstacles,

若這台車要突破這障礙物的阻擋，

the car has to detect them—

它首先得要能偵測到它們——

gleaning enough information about their size, shape, and position,

收集足夠的資訊，了解 它們的大小、形狀、位置，

so that its control algorithms can plot the safest course.

這麼一來，它的控制演算法 就能夠繪出最安全的路線。

With no human at the wheel,

沒有人在操作方向盤，

the car needs smart eyes, sensors that’ll resolve these details—

這台車需要有聰明的眼睛， 即能夠分析這些細節的感測器——

no matter the environment, weather, or how dark it is—

不論在什麼環境中、 什麼氣候下，不論天色有多黑——

all in a split-second.

都要瞬間判斷。

That’s a tall order, but there’s a solution that partners two things:

這簡直是苛求，但有個 結合兩種東西的解決方案：

a special kind of laser-based probe called LIDAR,

一種叫做雷射雷達的 特殊雷射探測器，

and a miniature version of the communications technology

以及讓網際網路能一直忙碌的

that keeps the internet humming, called integrated photonics.

一種迷你版本通訊技術， 叫做積體光學。

To understand LIDAR, it helps to start with a related technology— radar.

若要了解雷射雷達，應該先了解 一項相關技術——雷達。

In aviation,

在航空上，

radar antennas launch pulses of radio or microwaves at planes

雷達天線會向飛機發射 無線電波或微波的脈衝，

to learn their locations by timing how long the beams take to bounce back.

計算波束反彈回來的時間， 推算出飛機的位置。

That’s a limited way of seeing, though,

不過，這種看見的方式會受限，

because the large beam-size can’t visualize fine details.

因為大型波束無法 視覺化呈現精密的細節。

In contrast, a self-driving car’s LIDAR system,

相對的，自動駕駛汽車的 雷射雷達系統，

which stands for Light Detection and Ranging,

也就是「光學定向和測距」系統，

uses a narrow invisible infrared laser.

使用狹窄的不可見紅外線雷射。

It can image features as small as the button on a pedestrian’s shirt

它的成像能夠精密到 連對街行人的鈕扣

這種小特徵都不放過。

across the street.

But how do we determine the shape, or depth, of these features?

但，我們要如何決定 這些特徵的形狀或深度？

LIDAR fires a train of super-short laser pulses to give depth resolution.

雷射雷達會發射一連串 超短雷射脈衝來解析深度。

Take the moose on the country road.

以鄉下道路上的麋鹿為例。

As the car drives by, one LIDAR pulse scatters off the base of its antlers,

當汽車從旁邊開過時，

一個雷射雷達脈衝 會在它的鹿角基部散開，

while the next may travel to the tip of one antler before bouncing back.

而下一個脈衝則有可能會碰到 一支角的尖端，然後才反彈回來。

Measuring how much longer the second pulse takes to return

測量第二個脈衝花了 多少時間才彈回來，

provides data about the antler’s shape.

這樣的資料就能用來 判斷角的形狀。

With a lot of short pulses, a LIDAR system quickly renders a detailed profile.

雷射雷達用大量的短脈衝 便能快速提供出細節的側寫資訊。

The most obvious way to create a pulse of light is to switch a laser on and off.

若要創造脈衝光，最明顯的方式 就是把雷射開啟再關閉。

But this makes a laser unstable and affects the precise timing of its pulses,

但這會讓雷射不穩定， 且會影響到脈衝的時間精準度，

which limits depth resolution.

這就會限制了深度的解析度。

Better to leave it on,

最好是讓它一直開著，

and use something else to periodically block the light reliably and rapidly.

用其他的東西定期、 快速地阻擋光線。

That’s where integrated photonics come in.

這就是積體光學上場的時候了。

The digital data of the internet

網際網路的數位資料

is carried by precision-timed pulses of light,

由精確定時的脈衝光來傳輸，

some as short as a hundred picoseconds.

有些短到一百億分之一秒。 （0.000 000 000 1 秒）

One way to create these pulses is with a Mach-Zehnder modulator.

製造這類脈衝的方法之一， 就是用馬赫陳爾德干涉儀。

This device takes advantage of a particular wave property,

這個裝置會利用一種特殊的波特性，

called interference.

叫做干涉。

Imagine dropping pebbles into a pond:

想像把小卵石丟到池塘中：

as the ripples spread and overlap, a pattern forms.

漣漪散開和交疊時會形成圖案。

In some places, wave peaks add up to become very large;

在某些地方，波峰會加疊 在一起，變得非常大；

in other places, they completely cancel out.

在其他地方，它們則是完全抵銷。

The Mach-Zehnder modulator does something similar.

馬赫陳爾德干涉儀的做法很類似。

It splits waves of light along two parallel arms and eventually rejoins them.

它會沿著兩支平行的 臂桿把光波分開，

最終再將它們重新結合起來。

If the light is slowed down and delayed in one arm,

如果一支臂桿的光 被減緩下來並延遲，

the waves recombine out of sync and cancel, blocking the light.

重新結合的不同步光波 會彼此抵銷，使光線沒了。

By toggling this delay in one arm,

透過在一支臂桿製造延遲，

the modulator acts like an on/off switch, emitting pulses of light.

干涉儀的功能就變成 像是開關，放出脈衝光。

A light pulse lasting a hundred picoseconds

維持一百億分之一秒的脈衝光

leads to a depth resolution of a few centimeters,

能夠產生出幾公分的深度解析度，

but tomorrow’s cars will need to see better than that.

但未來的汽車需要 看得比那更清楚。

By pairing the modulator with a super- sensitive, fast-acting light detector,

把干涉儀和超級敏感、 反應快速的光偵測器搭配使用，

the resolution can be refined to a millimeter.

解析度可以精密到一公釐。

That’s more than a hundred times better

這比雙眼視力 1.0 的人

than what we can make out with 20/20 vision, from across a street.

看對街時能夠達到的解析度 還要清楚一百倍。

The first generation of automobile LIDAR has relied on complex spinning assemblies

第一代的汽車雷射雷達 要仰賴複雜的旋轉組件

that scan from rooftops or hoods.

從屋頂或是引擎蓋上做掃瞄。

With integrated photonics,

有了積體光學，

modulators and detectors are being shrunk to less than a tenth of a millimeter,

干涉儀和偵測器被縮到 小於十分之一公釐，

and packed into tiny chips that’ll one day fit inside a car’s lights.

裝載到小型晶片上， 將來可以放到車燈中。

These chips will also include a clever variation on the modulator

這些晶片還將包括 巧妙變造的干涉儀，

to help do away with moving parts and scan at rapid speeds.

得以擺脫可動的部件，以高速掃瞄。

By slowing the light in a modulator arm only a tiny bit,

這個額外的裝置 能將干涉儀的一支臂桿

this additional device will act more like a dimmer than an on/off switch.

只稍微減緩一點點， 它比較不像開關，更像調光器。

If an array of many such arms, each with a tiny controlled delay,

如果一長列這類 控制一點點延遲的臂桿

is stacked in parallel, something novel can be designed:

被平行堆疊起來， 就能設計出很新穎的東西：

a steerable laser beam.

可操控的雷射光束。

From their new vantage,

有了這種新優勢，

these smart eyes will probe and see more thoroughly

這些聰明的眼睛將能夠 探索和看得更完整，

than anything nature could’ve imagined—

勝過任何大自然想像得出的東西，

and help navigate any number of obstacles.

導航通過障礙物，不管多少。

All without anyone breaking a sweat—

無需流汗，毫不費力——

except for maybe one disoriented moose.

也許那隻迷惘的糜鹿除外。