Uber's Arizona Accident Raises Questions on Autonomous Vehicle Safety

On March 24, Tempe police cited a driver for making an illegal left turn and hitting one of Uber’s test Volvos while it was in self-driving mode.

by Mike Butler, East Valley Tribune (Mesa, Ariz.) / April 5, 2017

(TNS) -- Although Uber was not at fault for a bad automobile crash last Friday in Tempe, the incident opens up a window to examine just how safe the ride-hailing company is compared to its peers when it comes to autonomous driving.

Seeing self-driving Uber and Waymo vehicles has become routine for East Valley drivers on their daily commutes and lunchtime drives. The cars always have a driver ready to take control at any moment, plus a right-seat engineer who gathers data.

Nobody gives it much thought anymore.

Until something bad happens.

On March 24, Tempe police cited a driver for making an illegal left turn and hitting one of Uber’s test Volvos while it was in self-driving mode. The Uber SUV, occupied by a test driver and an engineer in the passenger seat, flipped on its side and a third car was struck. Police said nobody was seriously injured.

Uber temporarily suspended its self-driving program but was testing vehicles again by March 27.

In a Tempe police report released March 29, the driver who was cited and one witness told police that the Uber car appeared to speed up through the intersection of McClintock Drive and Don Carlos Avenue in order to beat a yellow light.

The Uber driver, who was not cited, said he was just entering the intersection when the light turned yellow.

"There was no time to react as there was a blind spot created by the line of traffic in the southbound left lane on McClintock," he said in his statement.

A few days before the Uber incident, a Tesla Model X operating on autopilot nudged a Phoenix police motorcycle at a stoplight. The Tesla braked for the red light, but then began moving forward slowly again, prompting the officer to get off the bike and move away. The Tesla bumped the fallen motorcycle, but caused no damage to either vehicle.

Waymo was involved in three minor Chandler accidents last year. Waymo drivers were manually operating the cars in two of the incidents. The third was rear-ended while operating autonomously.

In California, transportation officials measure the effectiveness of self-driving cars by how often drivers are forced to disengage from autonomous mode and take manual control. Companies that operate there are required to self-report disengagements.

Waymo, formerly known as the Google Self-Driving Car Project, is the leader of the pack by far on this metric. It has also been testing the longest, since 2009, and driven more than 2 million miles.

Waymo’s rate of disengages fell from 0.8 per 1,000 miles in 2015 to 0.2 in 2016. This was even as the company increased its driving by 50 percent in California, accumulating 635,868 miles. The number of disengages fell from 341 to 124. Waymo had no accidents in California in 2016.

Disengages occur primarily because of software glitches, according to Waymo, but they can also be caused by other drivers behaving erratically.

“This four-fold improvement reflects the significant work we’ve been doing to make our software and hardware more capable and mature,” said Dmitri Dolgov, head of Waymo’s self-driving technology, in a Feb. 1 blog post. “And because we’re creating a self-driving car that can take you from door to door, almost all of our time has been spent on complex urban or suburban streets … practicing advanced maneuvers such as making unprotected left turns and traversing multi-lane intersections.”

California didn’t have data on Uber because the company pulled out of the state and moved testing to Arizona after a standoff over reporting and permit requirements. It has since received a permit and has resumed testing in California; Uber tests in Pennsylvania in addition to Arizona.

However, the website Recode recently obtained internal Uber disengagement reports. For the week ending March 8, Uber cars could only go 0.8 miles on average before a driver had to take manual control. That would be 800 disengages per 1,000 miles. The fleet of 43 cars drives about 20,000 miles per week.

Recode also said the leaked documents showed that Uber cars averaged about 200 miles of autonomous driving before encountering a “critical” intervention, one that would have resulted in a car hitting a person or causing a significant property damage accident.

Uber didn’t respond to a request for comment.

Waymo is suing Uber, claiming that the ride-hailing company stole trade secrets critical to autonomous driving technology. Uber has denied the allegations.

©2017 East Valley Tribune (Mesa, Ariz.) Distributed by Tribune Content Agency, LLC.

NEW ON THE PODCAST

Tech Startups Can Help Revitalize Both Ends of Your Ballot