Fool's errand? Testing the safety of self-driving cars
Doing this much testing of autonomous cars would drive anyone crazy.
A new analysis finds that self-driving vehicles would have to be driven "hundreds of millions of miles" — and "sometimes hundreds of billions of miles" — to prove they do a better job at preventing fatalities and injuries than old-fashioned human-driving cars.
The analysis by Rand Corp. says that even under aggressive testing assumptions, it would takes tens of years, "sometimes hundreds of years," to actually drive that eye-popping amount of miles with existing fleets of self-driving cars.
And that, Rand says, is "an impossible proposition" if the goal is to demonstrate the relative safety of autonomous cars "prior to releasing them on the roads for consumer use."
"Our results show that developers of this technology and third-party testers cannot drive their way to safety," said Rand senior scientist Nidhi Kalra, co-author of the study.
"It's going to be nearly impossible for autonomous vehicles to log enough test-driving miles on the road to statistically demonstrate their safety when compared to the rate at which injuries and fatalities occur in human-controlled cars and trucks."
The analysis called for "innovative methods" to demonstrate the safety of self-driving cars, which could include accelerated testing, virtual testing and mathematical modeling. But it also warned that even if such methods are developed, "it may not be possible" to prove their safety prior to releasing them to the public.
"Uncertainty will persist," the study found.
The analysis comes three months after the Obama administration said it will, within the first half of 2016, "develop guidance on the safe deployment and operation of autonomous vehicles."
The administrator of the National Highway Traffic Safety Administration, Mark Rosekind, said in January that "NHTSA is using all of its available tools to accelerate the deployment of technologies that can eliminate 94 percent of fatal crashes involving human error."
"We will work with state partners toward creating a consistent national policy on these innovations, provide options now and into the future for manufacturers seeking to deploy autonomous vehicles, and keep our safety mission paramount at every stage."
The Rand study noted that self-driving vehicles have the potential "to significantly" reduce the 32,000 fatalities and more than 2 million injuries that result from auto crashes every year. More than 90 percent of those crashes are caused by human errors, the study points out.
"Autonomous vehicles are never drunk, distracted or tired," factors which are "involved in 41 percent, 10 percent and 2.5 percent of all fatal crashes, respectively," the study found.
And self-driving cars may be safer than human drivers in other regards, because they don't suffer from blind spots, and have better decision-making and better execution of steering, braking and acceleration, the analysis said.
But the study also said that "autonomous vehicles might not eliminate all crashes."
It noted that bad weather and "complex driving environments" present challenges for self-driving cars, and such vehicles "might perform worse than human drivers in some cases."
"There is also the potential for autonomous vehicles to pose new and serious crash risks," because of their exposure to cyberattacks, for example, the study said.
While self-driving vehicles offer "enormous potential benefits," they also present "enormous potential risks," the study said.
Rand senior statistician Susan Paddock, a co-author of the study, noted that "the most autonomous miles any developer has logged are about 1.3 million miles," which was done by Google's fleet of self-driving cars. And that "took several years," which "does not come close to the level of driving that is needed to calculate safety rates," Paddock said.
"Even if autonomous fleets are driving 10 million miles, one still would not be able to draw statistical conclusions about safety and reliability," Paddock said.