Why do driverless cars crash into each other? To save lives

One projection states that, in the fully autonomous future, on-road deaths will be reduced by 90%. The question is about what should be done in the interim, when self-driving cars share the road with human drivers and the illogical ways they can behave.

Even though many of the self-driving cars that are already in testing seem to be safer than the average human driver, there is still a lot of hesitation about rolling them out en masse. One hurdle is confronting the amount of negative attention that will be generated when an AV crashes. Hand-wringing about a crash caused by a self-driving car could set back self-driving cars tremendously.

The negative impact can even be seen in a recent accident involving an autonomous shuttle in Las Vegas.

The shuttle was operating on its first day when a truck driver started backing a vehicle towards the shuttle. The shuttle did what it was supposed to do and stopped, but the truck driver did not notice the shuttle and backed into its bumper. Even though the accident was entirely the fault of the human truck driver, most headlines about the incident focused solely on the autonomous shuttle. One can imagine the negative PR storm for an incident with graver consequences.

With these issues in mind, the question of when and how lawmakers should allow autonomous vehicles on our roads is a challenging one, and appears to be a sort of balancing act. Introducing imperfect AVs immediately could risk alienating consumers, but waiting until autonomous vehicles are 'perfect' would leave the issue of road death rates unaddressed.

The RAND Corporation addressed this scenario in a recent study called The Enemy of the Good: Estimating the Cost of Waiting for Nearly Perfect Automated Vehicles.

In the report, authors Nidhi Kalra and David G. Groves compared the costs and benefits of three hypothetical realities: Allowing AVs on the road when they're ten percent better than human drivers (Improve10), allowing them on the road when they're 75% better than human drivers (Improve75), and allowing 90% better (Improve90). The results validated those who favor getting them on our streets as soon as possible.

"We find that, in the short term, more lives are cumulatively saved under a more permissive policy (Improve10) than stricter policies requiring greater safety advancements (Improve75 or Improve90) in nearly all conditions, and those savings can be significant - hundreds of thousands of lives," Kalra and Groves wrote in a summary of their findings. "In the long term, more lives are cumulatively saved under an Improve10 policy than either Improve75 or Improve90 policies under all combinations of conditions we explored. In many cases, those savings can be more than half a million lives."

One reason behind the researchers' findings is the belief that it could be a very long time before self-driving cars are 75 to 90% safer than human drivers.

While the leading AV developers appear to have designed operating systems that can process and handle 99% of on-road situations, it's that remaining one percent - the unexpected, crazy and the abnormal - that can throw cars for a loop. These happen to be the scenarios that are most likely to lead to accidents, for AVs and humans alike. Many experts consider Level 5 autonomy to be an idea that will only exist theoretically in the near future. So just how close are today's cars to hitting the RAND Corporation's 'Improve10' numbers? It varies, but generally speaking they may not be as close as you might believe, according to Aarian Marshall at Wired.

"The 43 companies testing self-driving cars in California must submit public 'disengagement reports,' noting every time a human driver intervenes while behind the wheel of a self-driving car," Marshall wrote. "Last year's reports show these cars are getting better, but aren't all the way there: Waymo's cars averaged 5,128 miles between disengagements - pretty good! - while Mercedes-Benz did 1.8 - not so great. Today, autonomous vehicles are about as good as a standard crappy driver."

Typically, AV designers are the first to acknowledge that their products are far from perfect. But their work is driven by the strong belief that, one day, their work will help save thousands of lives every year. If Kalra and Groves are to be believed, that day ought to come sooner than later.