< Return to Video

How do self-driving cars "see"? - Sajan Saini

  • 0:08 - 0:15
    It’s late, pitch dark, and a self-driving
    car winds down a narrow country road.
  • 0:15 - 0:19
    Suddenly, three hazards appear
    at the same time.
  • 0:19 - 0:21
    What happens next?
  • 0:21 - 0:24
    Before it can navigate this
    onslaught of obstacles,
  • 0:24 - 0:26
    the car has to detect them—
  • 0:26 - 0:30
    gleaning enough information about
    their size, shape, and position,
  • 0:30 - 0:34
    so that its control algorithms
    can plot the safest course.
  • 0:34 - 0:36
    With no human at the wheel,
  • 0:36 - 0:41
    the car needs smart eyes, sensors
    that’ll resolve these details—
  • 0:41 - 0:44
    no matter the environment,
    weather, or how dark it is—
  • 0:44 - 0:46
    all in a split-second.
  • 0:46 - 0:50
    That’s a tall order, but there’s a
    solution that partners two things:
  • 0:50 - 0:54
    a special kind of laser-based probe
    called LIDAR,
  • 0:54 - 0:57
    and a miniature version of
    the communications technology
  • 0:57 - 1:01
    that keeps the internet humming,
    called integrated photonics.
  • 1:01 - 1:06
    To understand LIDAR, it helps to start
    with a related technology— radar.
  • 1:06 - 1:07
    In aviation,
  • 1:07 - 1:12
    radar antennas launch pulses
    of radio or microwaves at planes
  • 1:12 - 1:17
    to learn their locations by timing
    how long the beams take to bounce back.
  • 1:17 - 1:19
    That’s a limited way of seeing, though,
  • 1:19 - 1:23
    because the large beam-size
    can’t visualize fine details.
  • 1:23 - 1:26
    In contrast, a self-driving car’s
    LIDAR system,
  • 1:26 - 1:29
    which stands for Light Detection
    and Ranging,
  • 1:29 - 1:32
    uses a narrow invisible infrared laser.
  • 1:32 - 1:37
    It can image features as small as the
    button on a pedestrian’s shirt
  • 1:37 - 1:38
    across the street.
  • 1:38 - 1:42
    But how do we determine the shape,
    or depth, of these features?
  • 1:42 - 1:48
    LIDAR fires a train of super-short laser
    pulses to give depth resolution.
  • 1:48 - 1:51
    Take the moose on the country road.
  • 1:51 - 1:56
    As the car drives by, one LIDAR pulse
    scatters off the base of its antlers,
  • 1:56 - 2:01
    while the next may travel to the tip
    of one antler before bouncing back.
  • 2:01 - 2:04
    Measuring how much longer
    the second pulse takes to return
  • 2:04 - 2:07
    provides data about the antler’s shape.
  • 2:07 - 2:13
    With a lot of short pulses, a LIDAR system
    quickly renders a detailed profile.
  • 2:13 - 2:19
    The most obvious way to create a pulse
    of light is to switch a laser on and off.
  • 2:19 - 2:23
    But this makes a laser unstable and
    affects the precise timing of its pulses,
  • 2:23 - 2:26
    which limits depth resolution.
  • 2:26 - 2:27
    Better to leave it on,
  • 2:27 - 2:33
    and use something else to periodically
    block the light reliably and rapidly.
  • 2:33 - 2:36
    That’s where integrated photonics come in.
  • 2:36 - 2:38
    The digital data of the internet
  • 2:38 - 2:41
    is carried by precision-timed
    pulses of light,
  • 2:41 - 2:44
    some as short as a hundred picoseconds.
  • 2:44 - 2:49
    One way to create these pulses is
    with a Mach-Zehnder modulator.
  • 2:49 - 2:53
    This device takes advantage of a
    particular wave property,
  • 2:53 - 2:55
    called interference.
  • 2:55 - 2:58
    Imagine dropping pebbles into a pond:
  • 2:58 - 3:02
    as the ripples spread and overlap,
    a pattern forms.
  • 3:02 - 3:05
    In some places, wave peaks add
    up to become very large;
  • 3:05 - 3:08
    in other places, they completely
    cancel out.
  • 3:08 - 3:12
    The Mach-Zehnder modulator
    does something similar.
  • 3:12 - 3:17
    It splits waves of light along two
    parallel arms and eventually rejoins them.
  • 3:17 - 3:21
    If the light is slowed down and
    delayed in one arm,
  • 3:21 - 3:26
    the waves recombine out of sync and
    cancel, blocking the light.
  • 3:26 - 3:28
    By toggling this delay in one arm,
  • 3:28 - 3:34
    the modulator acts like an on/off switch,
    emitting pulses of light.
  • 3:34 - 3:36
    A light pulse lasting a hundred
    picoseconds
  • 3:36 - 3:40
    leads to a depth resolution of a
    few centimeters,
  • 3:40 - 3:43
    but tomorrow’s cars will need
    to see better than that.
  • 3:43 - 3:48
    By pairing the modulator with a super-
    sensitive, fast-acting light detector,
  • 3:48 - 3:51
    the resolution can be refined
    to a millimeter.
  • 3:51 - 3:53
    That’s more than a hundred times better
  • 3:53 - 3:57
    than what we can make out with
    20/20 vision, from across a street.
  • 3:57 - 4:03
    The first generation of automobile LIDAR
    has relied on complex spinning assemblies
  • 4:03 - 4:06
    that scan from rooftops or hoods.
  • 4:06 - 4:07
    With integrated photonics,
  • 4:07 - 4:13
    modulators and detectors are being shrunk
    to less than a tenth of a millimeter,
  • 4:13 - 4:18
    and packed into tiny chips that’ll one
    day fit inside a car’s lights.
  • 4:18 - 4:22
    These chips will also include a clever
    variation on the modulator
  • 4:22 - 4:27
    to help do away with moving parts
    and scan at rapid speeds.
  • 4:27 - 4:31
    By slowing the light in a modulator
    arm only a tiny bit,
  • 4:31 - 4:36
    this additional device will act more
    like a dimmer than an on/off switch.
  • 4:36 - 4:41
    If an array of many such arms, each with
    a tiny controlled delay,
  • 4:41 - 4:45
    is stacked in parallel, something novel
    can be designed:
  • 4:45 - 4:47
    a steerable laser beam.
  • 4:47 - 4:49
    From their new vantage,
  • 4:49 - 4:52
    these smart eyes will probe and
    see more thoroughly
  • 4:52 - 4:55
    than anything nature could’ve imagined—
  • 4:55 - 4:58
    and help navigate any number
    of obstacles.
  • 4:58 - 5:00
    All without anyone breaking a sweat—
  • 5:00 - 5:04
    except for maybe one disoriented moose.
Title:
How do self-driving cars "see"? - Sajan Saini
Speaker:
Sajan Saini
Description:

View full lesson: https://ed.ted.com/lessons/how-do-self-driving-cars-see-sajan-saini

It's late, pitch dark and a self-driving car winds down a narrow country road. Suddenly, three hazards appear at the same time. With no human at the wheel, the car uses smart eyes, sensors that'll resolve these details all in a split-second. How is this possible? Sajan Saini explains how LIDAR and integrated photonics technology make self-driving cars a reality.

Lesson by Sajan Saini, directed by Artrake Studio.

more » « less
Video Language:
English
Team:
closed TED
Project:
TED-Ed
Duration:
05:04
Elise Haadsma approved English subtitles for How do self-driving cars "see"?
Elise Haadsma accepted English subtitles for How do self-driving cars "see"?
lauren mcalpine edited English subtitles for How do self-driving cars "see"?

English subtitles

Revisions