Driver death rate: Zero

The case for self-driving cars boils down to one word: safety.

Since motor vehicles were popularized a century ago, they've brought mobility and liberty to hundreds of millions of people.

But this empowerment comes with a cost. Each year, about 35,000 people die in motor vehicles crashes on U.S. roads. In fact, more Americans have died in car crashes than in wars. But it's not just a problem in the U.S. -- 1.25 million people die in motor vehicle crashes each year, worldwide.

"Imagine you're 50 or 100 years into the future and looking back on society today, saying, 'What [did we do back then] that was absolutely barbaric, or absurd, or didn't make any sense?" Kyle Vogt, CEO of Cruise, the self-driving arm of General Motors told CNN Tech. "I think letting teenagers drive 5,000-pound cars at 80 mph while texting on their phones [would be on the list]."

Car crashes are overwhelmingly caused by human error, but fully autonomous vehicles promise to eliminate that element. Machines don't get distracted, text and drive, or drink and drive.

Thanks to sensors and cameras with 360-degree vision, self-driving car technology has the ability to predict, perceive and react to what happens on the road, such as a pedestrian crossing the street, more quickly than a human driver. It can also sense when other cars nearby are slowing and adjust its speed accordingly.

But before self-driving cars are allowed on roads without human supervision, the technology has to prove itself.

"Society wins when [the tech] is safer than a human driver," Chris Urmson, CEO of Aurora Innovation, a startup developing autonomous vehicle technology for automakers, told CNN Tech. "That really is the bar that we should be shooting to."

On U.S. roads, there's roughly one death for every 100 million miles traveled. A task for the industry is proving that the vehicles are safer than this number.

But autonomous vehicles aren't crash proof. In November, a self-driving shuttle from Navya, a French self-driving car company, was involved in a crash on its first day in service. Meanwhile, Uber self-driving SUV flipped on its side in Arizona last March. But neither vehicle was at fault as authorities concluded human drivers in other cars erred in judgment.

"How do we know that self-driving vehicles are that safe?" said Bryant Walker Smith, a professor at the University of South Carolina School of Law. "This is one of the really tough questions."

An autonomous vehicle might perform near flawlessly in one city, but struggle in another. For example, driving on a highway during a snowstorm is a far different challenge from cruising through a city's downtown while the sun sets, potentially blinding the car's cameras.

Autonomous vehicles will emerge first in selective areas where they've proven themselves. Waymo, for example, is launching its public trials in Phoenix, Arizona, where it has already tested extensively. Several companies are testing in Pittsburgh, which has varied terrain and four seasons.

During testing, companies track the statistical performance of self-driving cars on real roads and in computer simulations. With these simulations, engineers can create thousands of new challenges, and see how the vehicles perform. Waymo's vehicles, for example, have completed more than 2.5 billion miles of simulated driving throughout its testing in the last year.

"There's a lot of nuance in assessing the quality and safety of the driving -- it's not a single metric." Urmson said. "By looking at all of that data holistically, we're able to convince ourselves that we'd be comfortable with this out in the road, and then from there be able to convince both the regulatory bodies."