The Washington PostDemocracy Dies in Darkness

Watch a self-driving car learn to navigate narrow European streets like a human driver

Wayve claims their autonomous vehicle learns to drive through mistakes and imitation.

(Photo courtesy of Wayve)

Over more than a decade, self-driving vehicles have logged millions of miles on roadways across the globe.

Despite all that driving, researchers say, the machines are still unable to replicate the sophisticated problem-solving and spontaneity human drivers employ each time they get behind the wheel.

In their ambitious attempt to create an autonomous car service, companies like Waymo run their software through millions of potential scenarios, create three-dimensional maps using lasers, and outfit their vehicles powerful sensors like LIDAR that can cost more than the cars they guide.

The goal is to prepare the vehicle for anything it might encounter before it touches the road, creating a system of rules that predetermine behavior.

Now, an upstart British company called Wayve claims to have created a self-driving car using technology that almost sounds Stone Age compared to the competition.

“We’ve built a system which can drive like a human, using only cameras and a sat-nav,” the company said in a statement posted on their website this week. “This is only possible with end-to-end machine learning."

“No HD-maps, No expensive sensor/computer suite, No hand-coded rules, Driving on roads never-seen during training,” the statement adds.

Here’s what happens when police pull over a driverless car

The company released a video of their self-driving technology operating a vehicle through the narrow streets of Cambridge, England. The video shows a backup driver — who is able to take control of the vehicle at any time — with their hands lightly touching the steering wheel. The company claims the vehicle is autonomously navigating “complex, narrow urban European streets” it has never encountered before using a model based on only 20 hours of training data.

Using machine learning — a system in which algorithms are not hand-coded, but trained over time — the company claims its vehicles learn the same way human drivers do: through experience, mistakes, feedback and imitation. In effect, the company says, the car isn’t so much being taught how to drive, but being instructed how not to drive.

“Every time a safety driver intervenes or takes over control of the vehicle, we learn from that experience and that feedback,” Alex Kendall, the company’s co-founder and chief technology officer, says in the video. “And with each piece of data we’re able to train our system to get better and better and better.”

In the United States, companies like Waymo — formerly Google’s self-driving car project — have taken a much different approach to refining the self-driving experience. The company has a fleet of about 600 Chrysler Pacifica Hybrid vans equipped with navigation technology that includes “19 cameras (and) six LIDAR and radar sensors that can see an object the size of a soccer ball as far away as three football fields,” according to Fox News. Waymo has relied on backup drivers for much of its testing, though the company received a permit in California last year that allows it to test autonomous vehicles without a backup driver in the front seat, Reuters reported.

“Waymo is equipped with a lot more sensing capabilities,” Waymo CEO John Krafcik told the news outlet. “It can see chickens crossing the road."

How do you get people to trust autonomous vehicles? This company is giving them ‘virtual eyes.’

Last year, the company launched the nation’s first commercial self-driving taxi service in several suburbs outside Phoenix. The 24/7 service, dubbed Waymo One, lets customers summon self-driving minivans through a smartphone app like Uber or Lyft. So far, the program has been available to several hundred early riders, the company said, but there are plans to expand.

Walmart’s latest tool for ordering groceries: Google assistant

This teenager started playing video games 18 hours a day. Now he makes more money than most adults.

Boston Dynamics’ newest robot is a massive birdlike machine that works in a warehouse