The Uber Accident: Gauging the Fallout

The inevitable first pedestrian fatality attributable to a fully autonomous vehicle (AV), which occurred on 18 March in Tempe, Arizona, USA, predictably threw a number of companies in the nascent AV industry into damage-control mode. It also highlighted that many of the most vexing questions about the AV future lie outside of the realm of technology, in the squishier areas of liability, risk assessment and individual responsibility.

Uber in the hot seat

Most conspicuously in the hot seat is Uber, the company whose test vehicle, driving at 40 miles per hour in fully autonomous mode at night, with a backup human driver behind the wheel, crashed into a 49-year-old pedestrian walking her bicycle across the road.

Uber—a company that in recent years has had more than its share of bad public relations—was careful to express sympathy with the victim’s family and assert that it was “fully cooperating with local authorities in their investigation of the incident.” It also immediately pulled its test vehicles off the road in Tempe and three other cities, pending a full investigation. But some observers have argued that the pedestrian appeared to be walking outside of a marked crosswalk; to have shown inadequate caution and been oblivious to the oncoming vehicle; and to have moved into the field of view too quickly for human or car to react.

Complicating the story further was the release, by the Tempe Police Department, of a very troubling piece of video footage showing the moments leading up to the accident from the point of view of the vehicle, and also in dash-cam footage of the backup driver in the same interval. The road-view footage does indeed suggest that the pedestrian appeared suddenly in the field of view. But it also offers little evidence that the automated systems—which should have worked better at night than in the daytime, and have been able to sense objects not necessarily visible to a human driver—caused the car to slow down or take any evasive action.

Meanwhile, the backup driver, as depicted in the video, appeared to be paying little or no attention to the road for 10 of the 13 seconds covered by the video, and to have looked up from something in her lap only a fraction of a second before the crash occurred. Clearly, there is a lot of blame to go around.

Velodyne: Not lidar’s fault

The video will likely emerge as a key piece of evidence in the investigation, and will no doubt undergo Talmudic scrutiny by attorneys in the coming months. But in the near term the incident has thrown an unwelcome spotlight on Uber’s AV program, which has reportedly struggled to match the reliability ratings of its rivals.

For her part, Marta Thoma Hall, the president of Velodyne, which makes the lidar systems used in numerous AV programs, professed in an interview with Forbes to be “as baffled as anyone else” by why the car didn’t respond to the signal of a pedestrian crossing the road. Even so, she suggested that the accident probably wasn’t the lidar’s fault, and likely lay in the software systems that integrate and respond to input from lidar, radar, cameras and other onboard sensors.

Meanwhile, John Krafcik, the CEO of Waymo, the AV development arm of Google parent company Alphabet, reportedly expressed “a lot of confidence” that Waymo’s AV technology could have handled the situation that Uber’s car apparently couldn’t. (In early February, Waymo settled a high-profile, acrimonious lawsuit with Uber, tied to AV lidar, out of court.) Nonetheless, his remarks did reflect some concern that the incident could undermine public confidence in AVs.

Company reactions

There’s no evidence that Waymo plans to halt its own self-driving car testing in Tempe, which has been going on for more than a year. Ford and General Motors, two other major players, reportedly will also continue their public-road AV tests.

Indeed, only two companies beyond Uber have announced a suspension of their street-level test programs pending the outcome of the Uber investigation. One of those, the startup firm nuTonomy, temporarily halted its test program in Boston at the request of the city government. The other, Toyota, pulled its autonomous test cars from public roads two days after the accident, though it’s unclear how long they’ll remain grounded.

The long view

Notwithstanding that, the accident clearly hasn’t cooled Toyota’s interest in the long-term future of AVs. On the same day as the Japanese carmaker signaled the temporary halt in its public-road test program, the lidar firm Blackmore Sensors and Analytics announced a new round of US$18 million in funding from Toyota’s venture-finance arm (in partnership with BMW and several other firms).

Indeed, the Uber accident’s biggest impact on AV development may lie in the thicket of social and legal issues surrounding the advent of self-driving cars, rather than in the technology’s development. The crash highlights in particular the difficulty of assigning liability in such cases, and the need for the laws and the courts to adapt, possibly in ways that may prove uncomfortable to the public at large.

Above all, however, public acceptance of AVs may ultimately hinge on some subtle arguments regarding risk assessment—a task for which the average human is notoriously ill-equipped. In a 2017 report, the RAND corporation noted that of the more than 37,000 traffic-accident fatalities in the United States in 2016, some 90 percent were attributable to human error. And it stressed that in a practical sense, AV technology should not be held to an ideal standard, but to the more realistic bar of lowering that error rate. “Introducing autonomous vehicles when they are just better than human drivers—as opposed to nearly perfect—could save hundreds of thousands of lives over 30 years,” the report concluded.

Update: Corrected to remove language suggesting that Velodyne made the lidar systems used in Uber test cars. Neither Uber nor Velodyne has confirmed a relationship between the two parties. [26 March 2018, 16:30 EST]