13 near-misses? Google’s autonomous driving stats are no major setback

google car pulled over

Google's self-driving car fought the law, but the law won.

Credit: Zandr Milewski

You can't drive hundreds of thousands of miles and not have a few near misses. The goal is to reduce accidents.

RELATED TOPICS

A car pulls up to the curb at your hotel. There is no driver. You peak inside, and notice there is no steering wheel and no brakes or gas pedal. It looks like the car is completely robotic. Odd, huh? But would you get in and let the car drive you across town?

Most of us would say no at this point in the development of autonomous driving, the ambitious and slightly unnerving innovation that seems to spring up in the news almost every other day. Someday, we might not blink and climb aboard.

Recently, Google reported that there were 13 near-misses when testing their autonomous driving tech between September of 2015 and November of last year. In total, there were 272 mistakes as listed in a document submitted to the California DMV.

Those numbers are amazingly low considering Google has logged thousands of miles with their fleet of “ladybug” cars in the San Francisco area and their previous fleet. It might be in the neighborhood of 100,000 miles. This data suggests the self-driving revolution is more of an evolution because it reveals that there were 13 times in all of the testing in that period when the “robot” driver would have crashed. It feeds into the fear most everyday drivers have about letting the car take over completely.

And yet, I’m OK with the evolution. Every great engineering innovation takes small steps forward. Heck, I heard about that it took one company almost 1,700 iterations just to make a smart bra. I want a robo-car to think for me, and eventually there will be so many perks to autonomous driving that it will become commonplace. Here’s why.

First, remember that the infrastructure of self-driving does not exist -- unless you count a place called Mcity in Michigan, which is a small urban area where cars can connected to the stoplights and to other cars for testing purposes. My guess is that, if Google had logged all of those miles in an area that augmented the sensors in the car with connected traffic lights, sensors for the curbs and other obstructions, and with other cars, the near-misses would have been non-existent. A computer can scan in all directions at once, something a human driver can’t do; a computer can also connect to multiple sensors on the road at once.

We haven’t really tapped into the the benefits as much, either. While California requires that automakers like Audi and companies like Google report the near-misses, we don’t know how many times the computer saw something the human driver didn’t. The main benefit to autonomous driving is not that you can check your email while driving. It’s that the car can be 100% alert. Like speech recognition technology, the AI improves with each test and with each possible scenario. It only gets better. Google is accumulating a wealth of knowledge about what happens when a truck cuts into your path in a roundabout or someone gets a flat tire and swerves into your lane. Self-driving tech only improves over time; it doesn’t degrade.

There’s another really important factor here. Ford recently announced they had tested self-driving cars in winter (in Mcity). Audi drove an autonomous car from San Francisco to Las Vegas last year. Tesla already offers self-driving tech. It’s an industry wide project, and many companies are involved. What’s really missing is the support of a major automaker like Ford or Cadillac to take the work Google has done so far and make it ready for consumers. That’s partly a marketing effort, partly about integration, and mostly about safety. The automakers have shown an ability to test for safety in ways that Tesla obviously has not. Cadillac announced recently that they ran 50 million hours of simulations for the upcoming CT6 as part of their engineering effort. When Dodge tested the new Dart, they drove a fleet of them in a pre-production version for a total of a million miles. That’s pretty intense.

Maybe we’re not ready to let Google drive us home. But we should be ready to let autonomous driving tech provide extra safety to at least make our commutes less risky.

This article is published as part of the IDG Contributor Network. Want to Join?

RELATED TOPICS
Crash Course: Advanced beginner's guide to R
View Comments
Join the discussion
Be the first to comment on this article. Our Commenting Policies