Google's 14th Autonomous Car Accident — Still Not Its Fault


Pictured is the Google self-driving Lexus RX450h SUV. One of Google's retrofitted SUVs got in an accident on July 1. Image courtesy Google
Pictured is the Google self-driving Lexus RX450h SUV. One of Google's retrofitted SUVs got in an accident on July 1. Image courtesy Google

The Google autonomous-vehicle (AV) fleet saw its 14th accident this summer, and this time, there were injuries. Google said the other driver caused the crash, yet another instance of distracted driving on crowded city streets.

Google's AVs, or self-driving cars (it has nearly 50), use sensors to detect the car's surroundings, including cars, objects and people. Onboard software analyzes the sensor data to determine and enact the safest driving maneuvers. The prototype vehicles also can operate in manual mode when necessary.

Google reported that on July 1, one of its retrofitted Lexus SUVs was driving itself in rush-hour traffic in Mountain View, California, when it stopped at a green light. Why? The two cars ahead of it had stopped in order to avoid blocking the intersection, as there was no room for additional cars on the other side. The car traveling behind the AV then rear-ended it at about 17 mph (27 kph).

It wasn't a huge deal, according to Google. The AV suffered some scratches, the other car's front bumper fell off, and the passengers in both cars walked away. The Google employees in the AV suffered "minor" whiplash, and the other driver reported back and neck pain.

The company started self-reporting AV accidents in May in response to criticism that it was being too secretive (who, Google?) about the car's performance in testing. We now know that its autonomous vehicles have seen 14 collisions between 2010 and 2015. Only nine of those collisions took place while the car was actually driving itself, though. Nine crashes in five years sounds astronomical, but they took place over more than 1 million miles (1.6 million kilometers). That's about 77 years of driving for regular folk (assuming an average person drive 13,000 miles, or 20,922 kilometers, yearly). And anyway, all of the accidents were the other drivers' faults.

In March 2013, for instance, another vehicle veered into an AV's lane. In February 2015, another vehicle rolled through a stop sign. In April 2015, another AV was preparing to turn right on red, "creeping forward to obtain a better field of view of cross traffic," when its sensors detected an approaching car, so it stopped. At that time, the car behind it, which had also been creeping forward, "failed to brake sufficiently."

Google's confidence in its cars might be warranted. Of the nine collisions in autonomous mode, seven involved the AV being rear-ended; and in six of those, the AV was rightly at a complete stop when it was hit, as in the July 1 crash. In that one, Google explained, the driver of the other car obviously wasn't paying attention to the road. (Chris Urmson, head of Google's AV program, figured the driver was looking at a smartphone instead.)

In this case, Google produced a piece of hard evidence to support its blamelessness: sensor data from the AV, which appears to show that the other driver never hit the brakes, and there was time to do so. In Google's video showing the sensors' image of events, at least three seconds passed between the AV coming to a complete stop and the collision. The recommended following distance for safe braking is three seconds (from when the car in front of you passes a landmark to when you pass that landmark).

For Google, the collision adds to the growing pool of AV driving data that points to driver distraction as a significant menace on city streets. It also suggests its self-driving cars are, as suspected, a potentially viable cure. Google reportedly has distracted-driver-alert technology in the pipeline.