Skip to Content

Just how close to commercial reality is a self-driving car?

FORTUNE — Informed conversations about self-driving cars no longer are about feasibility. New key talking points are “When?” and “Which automakers first?” and “Who will be responsible when an accident happens?”

Nissan has said it will sell a driverless car by 2020. IHS forecast several models available by 2025. Both of these are guesses — but they indicate how fast the technology is progressing.

What seemed unimaginable a decade ago becomes more practical, comprehensible, and real by the day. Google’s (GOOG) self-driving Toyota Prius (TM) has logged hundreds of thousands of miles without incident on California roads. Most automakers are testing self-driving cars on tracks and — lately, as I experienced earlier this week in Las Vegas — in traffic.

MORE: Audi’s CES laser light show

The Audi A7 equipped with “traffic jam assist” was programmed to drive itself slowly in heavy traffic at no more than 40 miles per hour. (Dr. Bjorn Giesler, head of Audi’s project team, was behind the wheel.) The car was loaded with cameras, sensors, and a special device that monitors a driver’s eyes to ensure he or she doesn’t fall asleep at the wheel. In that event, the car will safely slow down, stop, and call for help.

Think of a driverless car as a robot. For Audi and other automakers, a key question is how much of the driving should be done by the robot, how much by the driver. The driver decides. Executives at Audi and other automakers say the driver, in any case, must remain engaged and attentive, ready to take over in the event of the unexpected: a car travelling the wrong way or out of control, for example.

Audi executives won’t use the word “driverless;” instead they speak about “piloted” driving. Other auto executives talk about “autonomous” or “assisted” driving. Only Google is adamant that it wants a driverless car, one that can help the elderly and the blind, as well as anyone who would rather be reading a book.

Today’s Audis and many other brands already may be equipped with features like adaptive cruise control that keeps a car a safe distance and constant speed behind cars ahead. Several have dynamic lane assist, which warn when a car is leaving a lane inadvertently — and can gently steer the car back.

Given a multitude of sensors, weather conditions, road changes, pedestrians, and other vehicles — robotic logic must be able to decide safely and instantaneously whether to turn, accelerate, or brake. The software, hardware, and algorithms that sift all this information are getting cheaper, smaller, and faster. Last year, the control systems filled the trunk of an Audi vehicle; this year, custom chips that function as the brain sit on a board about the size of a book.

MORE: How GM is making it near impossible to steal a Tahoe

State and federal regulators still must decide under what circumstances to permit so-called autonomous systems or, perhaps, whether to mandate features like adaptive cruise control or lane assist, if they are deemed to make automobile travel more safe.

Auto insurance today is a very straightforward process. But what about when a piloted car hits a pedestrian? Or when a truck hits a piloted car? Once piloted driving becomes more common, real-world experience will show how many accidents happen. Actuaries have statistical tools for assessing how much accidents will cost and, therefore, how much everyone will pay in insurance premiums.

As for legal responsibility, a question at an Audi press conference summed it up this way: “If a car without a driver has an accident, who is responsible: the driver? The owner of the car? Audi?” No one has answered the questions definitively, but it’s a good bet that driverless cars will be involved in far fewer accidents than ones with — otherwise, why have them?