Self-driving cars rely on high-definition maps to
know exactly where they are even if GPS fails.
Tesla collects data from its cars to build these maps
for its Autopilot and self-driving-car efforts.
Two former Tesla engineers are taking a page out of
Tesla’s book by crowdsourcing data to build maps other
automakers can use.
Andrew Kouri was working at Tesla when he noticed a big problem
standing in the way of self-driving cars.
When it comes to autonomous driving capabilities, Tesla’s cars
are among the most sophisticated on the road. The electric-car
maker’s Autopilot system allows the vehicles to drive in heavy
traffic and follow winding paths on highways.
But if the company wants its Model S and Model X cars to become
truly autonomous and capable of handling any driving scenario,
the vehicle’s cameras and sensors alone won’t be enough.
Instead, they will need something else — like high-definition
High-definition maps are like “sheet music” for self-driving
cars, Kouri, now a cofounder of the startup Lvl5, said in an
interview. The cars can rely on the maps and their different
landmarks to figure out where they are even when GPS fails.
Companies have struggled to get maps with the level of detail
necessary for self-driving cars. Not only do the maps have to be
highly detailed, but they also have to cover large swaths of land
and be regularly updated as road conditions change.
“It came as a shock to me when I was at Tesla that Tesla couldn’t
find anyone to buy these maps from because no one really makes
them yet,” Kouri said.
Lvl5 aims to address that issue.
The company made its debut at this year’s Y Combinator, an
accelerator program for startups. With $2 million in seed
funding, Lvl5 is crowdsourcing data with the goal of creating
high-definition maps of every route in the US. The startup aims
to sell its maps directly to car companies.
The Tesla effect
Lvl5 is taking a page right out of Tesla’s book.
In 2015, to address the mapping problem Kouri identified, Tesla
CEO Elon Musk decided the company should build its own maps by
collecting data from its Model S cars that were already on
the road. Tesla has since
ramped up those efforts by capturing short video clips from
cameras installed on both its Model S and Model X vehicles.
That data is being used to build maps that help Tesla vehicles
recognize things like street signs, traffic lights, and lane
Kouri and his Lvl5 cofounder Erik Reed were engineers on Tesla’s
internal mapping team. The two are now using a similar model with
Lvl5 has teamed up with Uber and Lyft drivers to collect data for
its maps. Kouri designed an app, called Payver, that is designed
to snap a picture of the road after every meter a driver travels.
To use it, drivers simply have to download the app and position
their phone so the device’s camera can view the road in front of
To help recruit Uber drivers, Kouri said, he went
window-to-window at a San Francisco International Airport parking
lot where Uber drivers wait to pick up passengers.
“I went in and just started knocking on windows, handing out free
phone mounts, and got a lot of people to sign up there,” he said.
The power of crowdsourcing
Since January, some 2,500 Uber and Lyft drivers have downloaded
the app, Kouri said. The startup has already mapped 500,000 miles
across the US.
Lvl5 is paying drivers $0.05 a mile when they take new routes.
Otherwise it pays them $0.02 a mile.
Anyone can try out the app, but Uber and Lyft drivers have more
of an incentive to use it because they drive enough to actually
see some significant money from it, Kouri said.
Lvl5’s biggest advantage over automakers like Tesla that are
trying to do the same thing is that its app can be used in any
car. That should give it wider coverage and allow it to collect
Tesla, by contrast, can collect data only from its own vehicles,
which limits the number of routes it can map.
Not all car manufacturers, Kouri said, “should have to have their
own mapping division.” He continued: “The goal is we will be the
mapping provider for all of these companies.”
But not every automaker interested in maps is going it alone.
BMW, Audi, and Mercedes, for example, are working together,
pooling their data and
handing it over to Here, a digital-map maker the three German
companies acquired for $3.1 billion in 2015.
Kouri said he considered Here and its rival navigation company
TomTom to be Lvl5’s biggest competitors. But because Lvl5 can
update its maps more regularly through crowdsourcing, it has an
advantage over them, he said.
“What we’re doing is taking this crowdsourcing approach that Waze
has and basically applying it to the self-driving-car problem,”
A drive for data
Waze, a real-time traffic app owned by Google, also collects data
from people’s smartphones to provide regular updates. The app,
however, doesn’t capture images of the road. Those images are
crucial for building maps for self-driving cars.
Waymo, Google’s self-driving sister company, uses vans equipped
with lidar to map routes. But lidar, a sophisticated sensor akin
to radar, is expensive. As a result, it’s installed on only a
small number of vehicles, and that limits how much mapping can be
done in a single day.
“If you only have 20 of these vans, there’s no way you’re going
to cover all the mileage in the world every day,” Kouri said.
Lvl5, however, doesn’t intend to rely on Payver in the long term,
Kouri said. Like Tesla and Here, the startup wants to collect
data captured by the cameras already installed on production
Unlike its rivals, though, Lvl5 doesn’t plan to enter any
exclusive partnerships, Kouri said. Instead, it wants to continue
to collect data from a wide array of car brands from all over the
It’s too early to tell whether the startup’s vision will pan out.
No automakers have signed up to use Lvl5’s mapping technology
yet, though one major automaker is testing it. Kouri declined to
name the automaker.
“If nobody has these global maps and nobody takes this
crowdsourcing approach like we are, then you and I won’t be able
to have self-driving cars take us from our door to work,” Kouri
Get the latest Tesla stock price here.