IIHS Study Finds Autonomous Vehicles Will Not Be the End of Crashes
Self-driving cars might only prevent a third of crashes
One of the biggest perks used to sell autonomous vehicles is the idea of collision-free roads, saving countless injuries and deaths every year. But while driver mistakes cause nearly all crashes each year, the Insurance Institute for Highway Safety suggests that even self-driving cars might only prevent a third of crashes. If they’re set to drive like people.
“It’s likely that fully self-driving cars will eventually identify hazards better than people, but we found that this alone would not prevent the bulk of crashes,” says Jessica Cicchino, IIHS vice president for research and a coauthor of the study.
The IIHS says that a US survey of police-reported crashes put driver error as the final failure leading to more than nine in 10 crashes, but their study showed that only around one-third of those would be avoided by the more accurate perception of self-driving cars and their lack of vulnerability to incapacitation.
“Building self-driving cars that drive as well as people do is a big challenge in itself,” says IIHS Research Scientist Alexandra Mueller, lead author of the study. “But they’d actually need to be better than [people] to deliver on the promises we’ve all heard.”
In the study, IIHS researchers examined more than 5,000 crashes from the National Motor Vehicle Crash Causation Survey collected by the National Highway Traffic Safety Administration, a representative sample of crashes where at least one vehicle was towed away and medical services were called.
They put the crashes into five categories, sensing and perceiving, covering errors like distraction and impeded visibility, predicting errors where drivers misjudged a gap or other vehicle speed, planning and deciding covering errors like short following distances and improper speed for conditions, execution and performance including inadequate evasive manoeuvres, and incapacitation which include impairment medical or fatigue-related crashes.
Crashes caused by sensing and perceiving errors and incapacitation made up 35 percent of crashes. Those, with perfectly functioning systems and sensors IIHS says, could be avoided with autonomous vehicles. Computer-driven cars are much less likely to misjudge speeds and distances, and they can’t be sleepy or under the influence. The remainder are more complicated.
IIHS uses the example of an Uber test vehicle that struck and killed a pedestrian in Arizona in 2018. The system struggled to identify the pedestrian, but when it did, the system did not predict her crossing and failed to take evasive action. An execution and performance error.
Planning and deciding errors, namely speeding and illegal manoeuvres, contributed to 40 percent of the crashes. To help avoid those, the IIHS says, vehicles would need to be programmed to prioritize safety over operator preference. The car would need to follow the rules, even when you didn’t want it to. To become safer than people, self-driving vehicles would also need to have strategies to account for uncertainty about other road users like driving more slowly with low visibility or in high pedestrian traffic situations where the speed limit doesn’t leave enough time to react to a dynamic situation.
“Our analysis shows that it will be crucial for designers to prioritize safety over rider preferences if autonomous vehicles are to live up to their promise to be safer than human drivers,” Mueller says. And a car that does what it thinks best over what you want it to could be a tough sell for many current drivers, especially those who put getting to their destination slightly quicker over the safety of others.