Skip to Content

They’re Racking up the Miles, but Are Self-Driving Cars Getting Safer?

Headlines about robot cars this week make the driverless revolution seem imminent, and yet still a long way off.
January 14, 2016

It’s a big week for robot cars. Transportation Secretary Anthony Foxx is in Detroit today to talk about how the Obama administration plans to smooth the path to getting autonomous cars on the road. The year 2020 keeps being reported as a year when that might happen, though that’s now less than four years away.

Google’s self-driving cars are going longer and longer before deciding to disengage and hand control back to humans.

There’s an in-depth piece posted on Medium today as well, looking at Google’s Self Driving Car Center in Atwater, California. Among other things, it introduces us to Google’s cadre of regular folks—among them a former baker, an artist, and a marketing exec—who the company employs as test (non)drivers.

Earlier this week, the Guardian looked at how safely these test drives are going. Google now has a fleet of 53 cars operating day and night, and they have logged 1.3 million miles of autonomous driving. Google is required by California state law to report “disengagements”—that is, times when either the computer has handed control of the car back to the driver, or the driver has taken over on his own. The reporting process isn’t entirely clear: the Medium piece says disengagements can happen anytime a car encounters something as simple as a piece of wood in the road and has to be steered around it. It suggests they happen several times per trip.

But according to the report obtained by the Guardian, Google reported just 341 disengagements from the end of 2014 through 2015. This low number may be because Google felt it didn’t have to report incidents that the car could have handled on its own.

Some data in the report do suggest the cars are getting better at driving on their own. The average distance between computer-initiated disengagements has gone down over time:

The report also says the trained drivers stepped in on average within 0.84 seconds of the computer signaling it was turning control over to a human.

That all sounds great, but each of these developments touches on perhaps the thorniest problem in self-driving car technology: the handoff between human and computer control. Trained test drivers like the ones Google employs may react quickly when the system needs help, but the research suggests that regular drivers are likely to zone out. Until cars improve to the point where they can match or exceed human performance in all situations, driver boredom may be the biggest problem holding back the robot car revolution.

Keep Reading

Most Popular

computation concept
computation concept

How AI is reinventing what computers are

Three key ways artificial intelligence is changing what it means to compute.

still from Embodied Intelligence video
still from Embodied Intelligence video

These weird virtual creatures evolve their bodies to solve problems

They show how intelligence and body plans are closely linked—and could unlock AI for robots.

seeing is believing concept
seeing is believing concept

Our brains exist in a state of “controlled hallucination”

Three new books lay bare the weirdness of how our brains process the world around us.

We reviewed three at-home covid tests. The results were mixed.

Over-the-counter coronavirus tests are finally available in the US. Some are more accurate and easier to use than others.

Stay connected

Illustration by Rose WongIllustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.