Self-driving cars are under intense scrutiny following the first pedestrian fatality involving a driverless vehicle, but some research suggests that getting autonomous vehicles on the road as soon as possible will save lives — even if they’re only slightly safer than human-piloted cars.
Uber halted testing of its self-driving vehicles after a 49-year-old woman was hit and killed in Tempe, Ariz., on Sunday, and Toyota TM, +1.10% has also stopped testing its driverless cars. “Our hearts go out to the victim’s family. We’re fully cooperating with @TempePolice and local authorities as they investigate this incident,” Uber wrote on Twitter.
Some observers say the tragedy will have negative repercussions across the new industry, and it will no doubt inform regulations that Congress is now debating. But some research suggests that slowing down the implementation of self-driving cars will do more harm than good. Cars that drive themselves some or most of the time only have to be slightly better drivers than humans to slash the rate of fatal car crashes, according to a November 2017 study by the RAND Corporation, a nonprofit nonpartisan think tank based in Santa Monica, Calif.
”One of the biggest questions on people’s minds is when should these vehicles be allowed on the road? There’s a lot of hand-wringing around this question and a lot of conjecture,” co-author Nidhi Kalra told MarketWatch in November. “We what we found surprised us.”
Self-driving cars will never crash because their drivers are drunk, distracted, sleepy, or speeding, but they have their own set of problems, such as attacks by hackers or hardware or software snafus, researchers noted.
And yet using computer modeling, RAND researchers compared three scenarios: one where driverless cars were just 10% safer than human drivers, one where they were 75% better and one where they were “nearly perfect,” or 90% better.
In both short (15 years) and long-term (30 years) scenarios, more lives were saved by quickly adopting the driverless cars that were just 10% better, researchers found. The difference was “significant,” ranging from tens of thousands to hundreds of thousands of saved lives, researchers found.
In one model, autonomous vehicles that were 10% safer were introduced to the marketplace in 2035 and saved 600,000 lives over the following 35 years. By 2070, there would be 1.5 million car crash fatalities if the driverless cars were on the road, compared with 2.1 million fatalities that would occur if they hadn’t been introduced, researchers found.
“We couldn’t find a single scenario in which it makes sense to wait until the vehicles are perfect,” Kalra said in November. “Waiting doesn’t save lives, relative to being a little more aggressive.”
One reason: once the cars are in use, their technology will quickly improve in response to exposure to real world scenarios that the cars’ developers wouldn’t foresee during simulations, Kalra said in November. One downside: There will undoubtedly be fatal crashes involving self-driving cars either way, which could create backlash against the technology. Kalra wasn’t available for comment immediately after the Tempe crash.
There are generally two schools of thought on when to allow driverless cars onto roads, Kalra said: Some feel the automated cars should hit the streets as soon as they are slightly better than human drivers, while others say the technology should be as close to perfect as possible.
There’s support for the former scenario: RAND’s study comes as auto fatalities have been on the rise. In 2016, 37,461 people were killed in car crashes, up from 35,092 in 2015, according to the National Highway Transportation Safety Administration.
Even so, drivers are only lukewarm about a driverless future. A 2016 survey by Kelley Blue Book found that most drivers (51%) would prefer to keep full control of their vehicle.
This story was originally published on Nov. 7, 2017. It was updated on March 20, 2018.