How a driverless car sees the road
-
0:01 - 0:04So in 1885, Karl Benz
invented the automobile. -
0:05 - 0:08Later that year, he took it out
for the first public test drive, -
0:08 - 0:12and -- true story --
crashed into a wall. -
0:12 - 0:14For the last 130 years,
-
0:14 - 0:19we've been working around that least
reliable part of the car, the driver. -
0:19 - 0:20We've made the car stronger.
-
0:20 - 0:23We've added seat belts,
we've added air bags, -
0:23 - 0:27and in the last decade, we've actually
started trying to make the car smarter -
0:27 - 0:30to fix that bug, the driver.
-
0:30 - 0:33Now, today I'm going to talk to you
a little bit about the difference -
0:33 - 0:37between patching around the problem
with driver assistance systems -
0:37 - 0:39and actually having fully
self-driving cars -
0:39 - 0:41and what they can do for the world.
-
0:41 - 0:44I'm also going to talk to you
a little bit about our car -
0:44 - 0:48and allow you to see how it sees the world
and how it reacts and what it does, -
0:48 - 0:51but first I'm going to talk
a little bit about the problem. -
0:52 - 0:53And it's a big problem:
-
0:53 - 0:561.2 million people are killed
on the world's roads every year. -
0:56 - 1:00In America alone, 33,000 people
are killed each year. -
1:00 - 1:02To put that in perspective,
-
1:02 - 1:07that's the same as a 737
falling out of the sky every working day. -
1:07 - 1:09It's kind of unbelievable.
-
1:10 - 1:12Cars are sold to us like this,
-
1:12 - 1:15but really, this is what driving's like.
-
1:15 - 1:17Right? It's not sunny, it's rainy,
-
1:17 - 1:19and you want to do anything
other than drive. -
1:19 - 1:21And the reason why is this:
-
1:21 - 1:23Traffic is getting worse.
-
1:23 - 1:26In America, between 1990 and 2010,
-
1:26 - 1:30the vehicle miles traveled
increased by 38 percent. -
1:30 - 1:33We grew by six percent of roads,
-
1:33 - 1:35so it's not in your brains.
-
1:35 - 1:39Traffic really is substantially worse
than it was not very long ago. -
1:39 - 1:41And all of this has a very human cost.
-
1:42 - 1:45So if you take the average commute time
in America, which is about 50 minutes, -
1:45 - 1:49you multiply that by the 120 million
workers we have, -
1:49 - 1:51that turns out to be
about six billion minutes -
1:51 - 1:53wasted in commuting every day.
-
1:53 - 1:56Now, that's a big number,
so let's put it in perspective. -
1:56 - 1:58You take that six billion minutes
-
1:58 - 2:02and you divide it by the average
life expectancy of a person, -
2:02 - 2:05that turns out to be 162 lifetimes
-
2:05 - 2:08spent every day, wasted,
-
2:08 - 2:10just getting from A to B.
-
2:10 - 2:12It's unbelievable.
-
2:12 - 2:14And then, there are those of us
who don't have the privilege -
2:14 - 2:16of sitting in traffic.
-
2:16 - 2:18So this is Steve.
-
2:18 - 2:19He's an incredibly capable guy,
-
2:19 - 2:22but he just happens to be blind,
-
2:22 - 2:25and that means instead of a 30-minute
drive to work in the morning, -
2:25 - 2:29it's a two-hour ordeal
of piecing together bits of public transit -
2:29 - 2:32or asking friends and family for a ride.
-
2:32 - 2:35He doesn't have that same freedom
that you and I have to get around. -
2:35 - 2:38We should do something about that.
-
2:38 - 2:40Now, conventional wisdom would say
-
2:40 - 2:42that we'll just take
these driver assistance systems -
2:42 - 2:46and we'll kind of push them
and incrementally improve them, -
2:46 - 2:48and over time, they'll turn
into self-driving cars. -
2:48 - 2:51Well, I'm here to tell you
that's like me saying -
2:51 - 2:55that if I work really hard at jumping,
one day I'll be able to fly. -
2:55 - 2:58We actually need to do
something a little different. -
2:58 - 3:00And so I'm going to talk to you
about three different ways -
3:00 - 3:04that self-driving systems are different
than driver assistance systems. -
3:04 - 3:06And I'm going to start
with some of our own experience. -
3:06 - 3:09So back in 2013,
-
3:09 - 3:11we had the first test
of a self-driving car -
3:11 - 3:13where we let regular people use it.
-
3:13 - 3:15Well, almost regular --
they were 100 Googlers, -
3:15 - 3:17but they weren't working on the project.
-
3:17 - 3:21And we gave them the car and we allowed
them to use it in their daily lives. -
3:21 - 3:25But unlike a real self-driving car,
this one had a big asterisk with it: -
3:25 - 3:26They had to pay attention,
-
3:26 - 3:29because this was an experimental vehicle.
-
3:29 - 3:32We tested it a lot,
but it could still fail. -
3:32 - 3:35And so we gave them two hours of training,
-
3:35 - 3:37we put them in the car,
we let them use it, -
3:37 - 3:39and what we heard back
was something awesome, -
3:39 - 3:41as someone trying
to bring a product into the world. -
3:41 - 3:43Every one of them told us they loved it.
-
3:43 - 3:47In fact, we had a Porsche driver
who came in and told us on the first day, -
3:47 - 3:49"This is completely stupid.
What are we thinking?" -
3:50 - 3:53But at the end of it, he said,
"Not only should I have it, -
3:53 - 3:56everyone else should have it,
because people are terrible drivers." -
3:57 - 3:59So this was music to our ears,
-
3:59 - 4:03but then we started to look at what
the people inside the car were doing, -
4:03 - 4:04and this was eye-opening.
-
4:04 - 4:07Now, my favorite story is this gentleman
-
4:07 - 4:11who looks down at his phone
and realizes the battery is low, -
4:11 - 4:15so he turns around like this in the car
and digs around in his backpack, -
4:15 - 4:17pulls out his laptop,
-
4:17 - 4:19puts it on the seat,
-
4:19 - 4:21goes in the back again,
-
4:21 - 4:24digs around, pulls out
the charging cable for his phone, -
4:24 - 4:27futzes around, puts it into the laptop,
puts it on the phone. -
4:27 - 4:29Sure enough, the phone is charging.
-
4:29 - 4:33All the time he's been doing
65 miles per hour down the freeway. -
4:33 - 4:36Right? Unbelievable.
-
4:36 - 4:39So we thought about this and we said,
it's kind of obvious, right? -
4:39 - 4:41The better the technology gets,
-
4:41 - 4:43the less reliable
the driver is going to get. -
4:43 - 4:46So by just making the cars
incrementally smarter, -
4:46 - 4:49we're probably not going to see
the wins we really need. -
4:49 - 4:53Let me talk about something
a little technical for a moment here. -
4:53 - 4:55So we're looking at this graph,
and along the bottom -
4:55 - 4:58is how often does the car
apply the brakes when it shouldn't. -
4:58 - 5:00You can ignore most of that axis,
-
5:00 - 5:03because if you're driving around town,
and the car starts stopping randomly, -
5:03 - 5:05you're never going to buy that car.
-
5:05 - 5:08And the vertical axis is how often
the car is going to apply the brakes -
5:08 - 5:11when it's supposed to
to help you avoid an accident. -
5:11 - 5:14Now, if we look at
the bottom left corner here, -
5:14 - 5:16this is your classic car.
-
5:16 - 5:19It doesn't apply the brakes for you,
it doesn't do anything goofy, -
5:19 - 5:21but it also doesn't get you
out of an accident. -
5:21 - 5:24Now, if we want to bring
a driver assistance system into a car, -
5:24 - 5:26say with collision mitigation braking,
-
5:26 - 5:29we're going to put some package
of technology on there, -
5:29 - 5:32and that's this curve, and it's going
to have some operating properties, -
5:32 - 5:35but it's never going to avoid
all of the accidents, -
5:35 - 5:37because it doesn't have that capability.
-
5:37 - 5:39But we'll pick some place
along the curve here, -
5:39 - 5:42and maybe it avoids half of accidents
that the human driver misses, -
5:42 - 5:44and that's amazing, right?
-
5:44 - 5:46We just reduced accidents on our roads
by a factor of two. -
5:46 - 5:50There are now 17,000 less people
dying every year in America. -
5:50 - 5:52But if we want a self-driving car,
-
5:52 - 5:55we need a technology curve
that looks like this. -
5:55 - 5:57We're going to have to put
more sensors in the vehicle, -
5:57 - 5:59and we'll pick some
operating point up here -
5:59 - 6:01where it basically never
gets into a crash. -
6:01 - 6:04They'll happen, but very low frequency.
-
6:04 - 6:06Now you and I could look at this
and we could argue -
6:06 - 6:10about whether it's incremental, and
I could say something like "80-20 rule," -
6:10 - 6:12and it's really hard to move up
to that new curve. -
6:12 - 6:15But let's look at it
from a different direction for a moment. -
6:15 - 6:19So let's look at how often
the technology has to do the right thing. -
6:19 - 6:22And so this green dot up here
is a driver assistance system. -
6:22 - 6:25It turns out that human drivers
-
6:25 - 6:28make mistakes that lead
to traffic accidents -
6:28 - 6:31about once every 100,000 miles in America.
-
6:31 - 6:34In contrast, a self-driving system
is probably making decisions -
6:34 - 6:38about 10 times per second,
-
6:38 - 6:39so order of magnitude,
-
6:39 - 6:42that's about 1,000 times per mile.
-
6:42 - 6:44So if you compare the distance
between these two, -
6:44 - 6:47it's about 10 to the eighth, right?
-
6:47 - 6:49Eight orders of magnitude.
-
6:49 - 6:51That's like comparing how fast I run
-
6:51 - 6:54to the speed of light.
-
6:54 - 6:57It doesn't matter how hard I train,
I'm never actually going to get there. -
6:57 - 7:00So there's a pretty big gap there.
-
7:00 - 7:04And then finally, there's how
the system can handle uncertainty. -
7:04 - 7:07So this pedestrian here might be
stepping into the road, might not be. -
7:07 - 7:10I can't tell,
nor can any of our algorithms, -
7:10 - 7:13but in the case of
a driver assistance system, -
7:13 - 7:15that means it can't take action,
because again, -
7:15 - 7:19if it presses the brakes unexpectedly,
that's completely unacceptable. -
7:19 - 7:22Whereas a self-driving system
can look at that pedestrian and say, -
7:22 - 7:24I don't know what they're about to do,
-
7:24 - 7:28slow down, take a better look,
and then react appropriately after that. -
7:28 - 7:31So it can be much safer than
a driver assistance system can ever be. -
7:31 - 7:34So that's enough about
the differences between the two. -
7:34 - 7:37Let's spend some time talking about
how the car sees the world. -
7:37 - 7:39So this is our vehicle.
-
7:39 - 7:41It starts by understanding
where it is in the world, -
7:41 - 7:44by taking a map and its sensor data
and aligning the two, -
7:44 - 7:47and then we layer on top of that
what it sees in the moment. -
7:47 - 7:51So here, all the purple boxes you can see
are other vehicles on the road, -
7:51 - 7:53and the red thing on the side
over there is a cyclist, -
7:53 - 7:55and up in the distance,
if you look really closely, -
7:55 - 7:57you can see some cones.
-
7:57 - 8:00Then we know where the car
is in the moment, -
8:00 - 8:04but we have to do better than that:
we have to predict what's going to happen. -
8:04 - 8:07So here the pickup truck in top right
is about to make a left lane change -
8:07 - 8:10because the road in front of it is closed,
-
8:10 - 8:11so it needs to get out of the way.
-
8:11 - 8:13Knowing that one pickup truck is great,
-
8:13 - 8:16but we really need to know
what everybody's thinking, -
8:16 - 8:18so it becomes quite a complicated problem.
-
8:18 - 8:23And then given that, we can figure out
how the car should respond in the moment, -
8:23 - 8:27so what trajectory it should follow, how
quickly it should slow down or speed up. -
8:27 - 8:30And then that all turns into
just following a path: -
8:30 - 8:33turning the steering wheel left or right,
pressing the brake or gas. -
8:33 - 8:35It's really just two numbers
at the end of the day. -
8:35 - 8:38So how hard can it really be?
-
8:38 - 8:40Back when we started in 2009,
-
8:40 - 8:42this is what our system looked like.
-
8:42 - 8:46So you can see our car in the middle
and the other boxes on the road, -
8:46 - 8:47driving down the highway.
-
8:47 - 8:51The car needs to understand where it is
and roughly where the other vehicles are. -
8:51 - 8:53It's really a geometric
understanding of the world. -
8:53 - 8:56Once we started driving
on neighborhood and city streets, -
8:56 - 8:58the problem becomes a whole
new level of difficulty. -
8:58 - 9:02You see pedestrians crossing in front
of us, cars crossing in front of us, -
9:02 - 9:04going every which way,
-
9:04 - 9:05the traffic lights, crosswalks.
-
9:05 - 9:08It's an incredibly complicated
problem by comparison. -
9:08 - 9:10And then once you have
that problem solved, -
9:10 - 9:13the vehicle has to be able
to deal with construction. -
9:13 - 9:16So here are the cones on the left
forcing it to drive to the right, -
9:16 - 9:18but not just construction
in isolation, of course. -
9:18 - 9:22It has to deal with other people moving
through that construction zone as well. -
9:22 - 9:25And of course, if anyone's
breaking the rules, the police are there -
9:25 - 9:29and the car has to understand that
that flashing light on the top of the car -
9:29 - 9:32means that it's not just a car,
it's actually a police officer. -
9:32 - 9:34Similarly, the orange box
on the side here, -
9:34 - 9:35it's a school bus,
-
9:35 - 9:38and we have to treat that
differently as well. -
9:39 - 9:41When we're out on the road,
other people have expectations: -
9:41 - 9:43So, when a cyclist puts up their arm,
-
9:43 - 9:47it means they're expecting the car
to yield to them and make room for them -
9:47 - 9:49to make a lane change.
-
9:49 - 9:51And when a police officer
stood in the road, -
9:51 - 9:54our vehicle should understand
that this means stop, -
9:54 - 9:57and when they signal to go,
we should continue. -
9:57 - 10:01Now, the way we accomplish this
is by sharing data between the vehicles. -
10:01 - 10:03The first, most crude model of this
-
10:03 - 10:05is when one vehicle
sees a construction zone, -
10:05 - 10:08having another know about it
so it can be in the correct lane -
10:08 - 10:10to avoid some of the difficulty.
-
10:10 - 10:12But we actually have a much
deeper understanding of this. -
10:12 - 10:15We could take all of the data
that the cars have seen over time, -
10:15 - 10:18the hundreds of thousands
of pedestrians, cyclists, -
10:18 - 10:19and vehicles that have been out there
-
10:19 - 10:21and understand what they look like
-
10:21 - 10:24and use that to infer
what other vehicles should look like -
10:24 - 10:26and other pedestrians should look like.
-
10:26 - 10:29And then, even more importantly,
we could take from that a model -
10:29 - 10:31of how we expect them
to move through the world. -
10:31 - 10:34So here the yellow box is a pedestrian
crossing in front of us. -
10:34 - 10:37Here the blue box is a cyclist
and we anticipate -
10:37 - 10:40that they're going to nudge out
and around the car to the right. -
10:40 - 10:42Here there's a cyclist
coming down the road -
10:42 - 10:46and we know they're going to continue
to drive down the shape of the road. -
10:46 - 10:48Here somebody makes a right turn,
-
10:48 - 10:51and in a moment here, somebody's
going to make a U-turn in front of us, -
10:51 - 10:54and we can anticipate that behavior
and respond safely. -
10:54 - 10:56Now, that's all well and good
for things that we've seen, -
10:56 - 10:59but of course, you encounter
lots of things that you haven't -
10:59 - 11:00seen in the world before.
-
11:00 - 11:02And so just a couple of months ago,
-
11:02 - 11:04our vehicles were driving
through Mountain View, -
11:04 - 11:06and this is what we encountered.
-
11:06 - 11:08This is a woman in an electric wheelchair
-
11:08 - 11:11chasing a duck in circles on the road.
(Laughter) -
11:11 - 11:14Now it turns out, there is nowhere
in the DMV handbook -
11:14 - 11:16that tells you how to deal with that,
-
11:16 - 11:18but our vehicles were able
to encounter that, -
11:18 - 11:20slow down, and drive safely.
-
11:20 - 11:22Now, we don't have to deal
with just ducks. -
11:22 - 11:26Watch this bird fly across in front of us.
The car reacts to that. -
11:26 - 11:28Here we're dealing with a cyclist
-
11:28 - 11:31that you would never expect to see
anywhere other than Mountain View. -
11:31 - 11:33And of course, we have
to deal with drivers, -
11:33 - 11:37even the very small ones.
-
11:37 - 11:41Watch to the right as someone
jumps out of this truck at us. -
11:42 - 11:45And now, watch the left as the car
with the green box decides -
11:45 - 11:49he needs to make a right turn
at the last possible moment. -
11:49 - 11:52Here, as we make a lane change,
the car to our left decides -
11:52 - 11:55it wants to as well.
-
11:55 - 11:58And here, we watch a car
blow through a red light -
11:58 - 12:00and yield to it.
-
12:00 - 12:04And similarly, here, a cyclist
blowing through that light as well. -
12:04 - 12:07And of course,
the vehicle responds safely. -
12:07 - 12:09And of course, we have people
who do I don't know what -
12:09 - 12:13sometimes on the road, like this guy
pulling out between two self-driving cars. -
12:13 - 12:15You have to ask, "What are you thinking?"
-
12:15 - 12:16(Laughter)
-
12:16 - 12:19Now, I just fire-hosed you
with a lot of stuff there, -
12:19 - 12:21so I'm going to break one of these
down pretty quickly. -
12:21 - 12:24So what we're looking at is the scene
with the cyclist again, -
12:24 - 12:28and you might notice in the bottom,
we can't actually see the cyclist yet, -
12:28 - 12:30but the car can: it's that little
blue box up there, -
12:30 - 12:32and that comes from the laser data.
-
12:32 - 12:35And that's not actually
really easy to understand, -
12:35 - 12:38so what I'm going to do is I'm going
to turn that laser data and look at it, -
12:38 - 12:41and if you're really good at looking
at laser data, you can see -
12:41 - 12:43a few dots on the curve there,
-
12:43 - 12:45right there, and that blue box
is that cyclist. -
12:45 - 12:46Now as our light is red,
-
12:46 - 12:49the cyclist's light
has turned yellow already, -
12:49 - 12:51and if you squint, you can see that
in the imagery. -
12:51 - 12:54But the cyclist, we see, is going
to proceed through the intersection. -
12:54 - 12:57Our light has now turned green,
his is solidly red, -
12:57 - 13:01and we now anticipate that this bike
is going to come all the way across. -
13:01 - 13:05Unfortunately the other drivers next to us
were not paying as much attention. -
13:05 - 13:08They started to pull forward,
and fortunately for everyone, -
13:08 - 13:11this cyclists reacts, avoids,
-
13:11 - 13:13and makes it through the intersection.
-
13:13 - 13:15And off we go.
-
13:15 - 13:18Now, as you can see, we've made
some pretty exciting progress, -
13:18 - 13:20and at this point we're pretty convinced
-
13:20 - 13:22this technology is going
to come to market. -
13:22 - 13:26We do three million miles of testing
in our simulators every single day, -
13:26 - 13:29so you can imagine the experience
that our vehicles have. -
13:29 - 13:32We are looking forward to having
this technology on the road, -
13:32 - 13:35and we think the right path
is to go through the self-driving -
13:35 - 13:37rather than driver assistance approach
-
13:37 - 13:39because the urgency is so large.
-
13:39 - 13:42In the time I have given this talk today,
-
13:42 - 13:4534 people have died on America's roads.
-
13:45 - 13:47How soon can we bring it out?
-
13:47 - 13:51Well, it's hard to say because
it's a really complicated problem, -
13:51 - 13:53but these are my two boys.
-
13:53 - 13:57My oldest son is 11, and that means
in four and a half years, -
13:57 - 13:59he's going to be able
to get his driver's license. -
13:59 - 14:03My team and I are committed
to making sure that doesn't happen. -
14:03 - 14:04Thank you.
-
14:04 - 14:08(Laughter) (Applause)
-
14:09 - 14:12Chris Anderson: Chris,
I've got a question for you. -
14:12 - 14:14Chris Urmson: Sure.
-
14:14 - 14:18CA: So certainly, the mind of your cars
is pretty mind-boggling. -
14:18 - 14:23On this debate between
driver-assisted and fully driverless -- -
14:23 - 14:26I mean, there's a real debate
going on out there right now. -
14:26 - 14:29So some of the companies,
for example, Tesla, -
14:29 - 14:31are going the driver-assisted route.
-
14:31 - 14:36What you're saying is that
that's kind of going to be a dead end -
14:36 - 14:42because you can't just keep improving
that route and get to fully driverless -
14:42 - 14:45at some point, and then a driver
is going to say, "This feels safe," -
14:45 - 14:48and climb into the back,
and something ugly will happen. -
14:48 - 14:50CU: Right. No, that's exactly right,
and it's not to say -
14:50 - 14:54that the driver assistance systems
aren't going to be incredibly valuable. -
14:54 - 14:56They can save a lot of lives
in the interim, -
14:56 - 15:00but to see the transformative opportunity
to help someone like Steve get around, -
15:00 - 15:02to really get to the end case in safety,
-
15:02 - 15:04to have the opportunity
to change our cities -
15:04 - 15:09and move parking out and get rid of
these urban craters we call parking lots, -
15:09 - 15:10it's the only way to go.
-
15:10 - 15:12CA: We will be tracking your progress
with huge interest. -
15:12 - 15:17Thanks so much, Chris.
CU: Thank you. (Applause)
- Title:
- How a driverless car sees the road
- Speaker:
- Chris Urmson
- Description:
-
Statistically, the least reliable part of the car is ... the driver. Chris Urmson heads up Google's driverless car program, one of several efforts to remove humans from the driver's seat. He talks about where his program is right now, and shares fascinating footage that shows how the car sees the road and makes autonomous decisions about what to do next.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 15:29
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast approved English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road | ||
Morton Bast edited English subtitles for How a driverless car sees the road |