TECH

What went wrong with Uber's Volvo in fatal crash? Experts shocked by technology failure

The Republic | azcentral.com

An Uber vehicle cruises in Tempe, Aug. 25, 2017, near Mill Avenue and Rio Salado Parkway.

Self-driving car experts are not surprised that someone was killed in a pedestrian accident involving the cutting-edge technology — nobody thought the cars would prevent all accidents.

But they are shocked at how dramatically the technology failed.

A 49-year-old woman was killed Sunday night in Tempe when she jaywalked with her bike across a dark section of Mill Avenue south of Curry. A self-driving Volvo operated by Uber struck and killed her.

It was exactly the type of accident Uber and other self-driving car companies say their advanced sensors and computer programming should prevent.

"This is not something I would expect artificial intelligence to struggle with," said Jim McPherson, a California attorney and self-driving car consultant who has criticized the industry for moving too fast to put untested technology on public roads.

Video released Wednesday of the accident shows the car not appearing to brake or steer away from pedestrian Elaine Herzberg as she walked across an open lane and in front of the car.

Herzberg appeared to be looking away from the oncoming vehicle, while an in-car camera shows Uber driver Rafaela Vasquez looking down at something below the dashboard, out of view of the camera, before the collision.

Drivers remain behind the wheel of Uber's cars to take over when the cars' autonomous systems fail.

The National Transportation Safety Board and National Highway Traffic Safety Administration are investigating, and industry experts are eager to see what they find regarding the robot car's performance in what should have been a routine interaction with a pedestrian.

McPherson said allowing companies to test autonomous cars on public roadways should be re-examined in light of the accident, and testing perhaps should be limited to private proving grounds until the companies can statistically show the cars are safe.

"They should all definitely reassess," McPherson said.

Possible programming failure

McPherson and other experts also question whether there was a problem with how the car was programmed.

It's possible that the sensors in the car did detect the pedestrian and the car "decided" not to act based on its programming, said James Arrowood, who runs Arrowood Attorneys in Scottsdale and teaches the State Bar of Arizona course on driverless cars.

"The computer makes a decision. It says, 'Hey, there is this object moving 10 or 15 feet to left of me, do I move or not?' " he said. "It (could be) programmed, I have a right of way, on the assumption that whatever is moving will yield the right of way."

Arrowood is an expert in motor-vehicle and product-liability issues. He based his comments on the public video only and does not represent anyone involved in the accident.

"Really what we are going to ask is, at what point should or could those sensors recognize the movement off to the left," he said. "Presumably she was somewhere in the darkness."

He said it is prudent for Uber to keep its cars off the road until the company understands why the car failed to, apparently, attempt to avoid the collision.

"What they are going to do is specifically look at the sensitivity of the sensors at night," he said. "I know from representing vehicle manufacturers, they take this stuff seriously. They don't want someone injured with it."

Drivers say sensors detect jaywalkers

Uber has operated its vehicles in the area for more than a year, and its dozens of Volvo SUVs have logged substantial miles on Mill Avenue where the accident occurred.

The company offered a reporter and photographer from The Arizona Republic rides in its Volvos last fall and operators told stories of when the sensor technology was more effective than human drivers.

The Uber employees who spend full-day shifts behind the wheel of the cars said that at night on Mill — when numerous people cross the street amid the restaurants and bars near Arizona State University — the cars are effective at anticipating jaywalkers.

The operators said in the desert area to the north, near the accident scene, the sensors often prompt the cars to stop before operators see pedestrians who are walking in the dark on the side of the road toward the path of the vehicle.

However, during a daylight drive around ASU and up and down Scottsdale Road, the operators of the Uber vehicles had to take over the controls for multiple instances when the cars could not negotiate traffic on their own.

In one instance, the driver braked and turned the wheel because he was concerned the autonomous Volvo would not avoid a car in front of it as the Volvo changed lanes to turn onto Rio Salado Parkway.

Uber has touted the self-driving program as a way to improve safety.

"Self-driving cars have the potential to save millions of lives and improve quality of life for people around the world," Uber's press material for the program says. "1.3 million people die every year in car accidents — 94 percent of those accidents involve human error. In the future we believe this technology will mean less congestion, more affordable and accessible transportation, and far fewer lives lost in car accidents."

This image shows the three dimensional image the sensors on an autonomous car operated by Uber generated Oct. 26, 2017 to negotiate traffic on Scottsdale Road.

How the cars 'see'

Uber's cars use a combination of cameras, radar and lidar to navigate and avoid obstacles. Lidar is like radar but using lasers instead of radio waves to construct a three-dimensional representation of the car's surroundings. Uber officials have said it can detect objects 100 yards away.

"Lidar or radar, each has their shortcomings depending on light, reflectivity," McPherson said. "Dark clothing and metal appear to one better than the other."

Uber customers who have used the self-driving cars in the last year in Tempe have seen the three-dimensional images the sensors create. Uber installed special screens in the rear passenger area to explain the technology to riders.

The screens show the car and nearby curbs, landscaping, and as many as three cars ahead. The detail is pixelated, like an old video game, but the sensors show enough detail to pick out passengers riding in a bus next to the Uber vehicle.

No big payout likely for victim

Regardless of the failure of the car's technology, Arrowood said the accident appears to be primarily caused by the pedestrian, and anyone suing for damages on her behalf faces a challenge.

Law firm Bellah Perez on Thursday announced it is representing Herzberg's daughter, though officials did not say explicitly they will sue Uber or the driver.

"A lot of people think (Uber) will just write a check for a billion dollars or something," Arrowood said. "That is not going to happen."

He said he expects a settlement, not a jury trial. But any settlement Uber might pay will be calculated based on the victim's lifetime earnings and life expectancy, both presumably low for a person living on the street.

"Whether people think that is fair or not, there is an analysis of lifetime earnings that is going to temper any judgement in a case like this," he said. "People don't like that. They say life is worth an equal amount. It is, but not under the law necessarily."

A question for jurors, if it goes that route, will be whether the robot car should have a higher standard than a human driver, he said.

He said the video shows a driver would have had about 1 second to react after the pedestrian came into view, and that it takes drivers about two seconds on average to spot and react to such an obstacle.

"You would have a difficult time trying to convince a jury the outcome would have been different had it just been a driver in the vehicle," he said.

Pedestrian can be blamed, technology notwithstanding

When pedestrians are in a crosswalk and obeying traffic signals, drivers have the responsibility to not hit them, he said. But when pedestrians cross outside of a crosswalk, it is their responsibility to yield to traffic and not get hit.

That makes the pedestrian negligent in this accident, he said.

"The company has a good defense," Arrowood said of Uber. "The video helps them."

He said the fact that the driver is looking away from the road at something below the dashboard area isn't great for the Uber.

"There will be some questions as to what the safety driver was doing and should have been doing," he said. "A smart attorney will go after Uber for the training of the safety driver."

The law firm representing Herzberg's daughter mentioned that Arizona has welcomed this technology as a reason the case is important, saying it has shifted accident liability in self-driving car accidents.

“As an Arizona law firm, we feel a special responsibility to represent this case as it directly impacts our fellow Arizonans sharing the road with these machines,” partner Cristina Perez Hesano said.

Arrowood said that because of Gov. Doug Ducey's 2015 executive order that permits testing of self-driving cars, a defense attorney for Uber could ask a judge to prevent a jury from considering the fact that the car was only on the road for research purposes.

"That would be prejudicial against the company," he said.

Governor: Put safety first

Ducey made only an oblique reference to the self-driving Uber accident in a Thursday speech at the Arizona Tech Innovation Summit.

“I would emphasize that with any new technology, whether it is self-driving vehicles, drones or anything else, we need to make sure we are putting public sector safety first,” he said. “That’s what Arizona has been doing and it will continue to do.”

Ducey told the crowd about how he attracted the ride sharing industry to Arizona in 2015. He said a state agency wanted to shut down companies like Uber just before the Phoenix area hosted the Super Bowl.

“Instead, we shuttered the agency,” he said to applause.

Arizona Republic reporter Richard Ruelas contributed to this story.