Robotic cars are fast taking ‘autopilot’ to new levels

Times Staff Writer

The predictions of futurists have often fizzled on the subject of robots, which today can vacuum floors and play chess but not drive a car.

But an exciting demonstration several weeks ago in the Nevada desert suggests that technologists are getting closer than anybody realized to a robotic car.

Within about two years, the first car able to autonomously drive on freeways will be a reality, predicts Sebastian Thrun, Stanford University’s guru of robotic cars and the winner of the Pentagon’s Grand Challenge race in October.

The Grand Challenge, sponsored by the Defense Advanced Research Projects Agency, pitted teams that had built cars able to autonomously navigate and drive an off-road course in the Nevada desert.

The Stanford team won a $2-million prize for completing the course in the shortest time among a field of 23 finalists, five of which were able to cross the finish line. In a similar contest last year, not a single entry finished.


Thrun admits he is not much of an expert on cars, although he is director of one of a handful of artificial-intelligence labs nationwide that are directing serious attention to developing cars that can drive themselves.

“I am a big fan of putting the intelligence in the cars,” Thrun says.

That statement marks a shift in thinking, coming after decades and billions of dollars in government spending on intelligent highways. The Bush administration has sharply increased such federal outlays, which have reached hundreds of millions of dollars a year.

The Transportation Department has poured this money into programs to install instruments on roadways and create intelligent networks that could operate fleets of vehicles. The investments have been much like federal research into fusion energy, with breakthroughs predicted as being just around the corner but never quite happening.

At one time, futurists envisioned massive networks of computers running society, but they missed foreseeing the personal computer and the Internet. The ability of individual cars to drive themselves without an intelligent highway network may represent the same conceptual mistake.

The Grand Challenge results this year were a real breakthrough, demonstrating that individual cars could successfully use satellite guidance, artificial vision and complex software to navigate around obstacles, away from ruts and through tunnels.

Stanford spent about $500,000 on its car and received another $500,000 in assistance from its partner Volkswagen.

The car, named Stanley, was equipped with a global positioning system, a series of laser range finders and a video camera, all connected to a computer that made decisions about how to navigate the course.

The vehicle completed the 131-mile course at an average speed of 19 miles per hour.

No doubt a human driver could have beaten the car’s time, because people can still handle a steering wheel more adeptly than a computer can. But perhaps not for much longer. For decades, the best chess players could beat computers, but no more.

Thrun is unrestrained in his enthusiasm for the technology, saying the challenge of navigating the off-road course -- with vegetation, ruts and rocks -- was greater than keeping a car on a paved highway. Of course, an urban course would be full of moving obstacles, many operated by maniacs.

Nonetheless, Thrun hopes that within two years his team will be able to build a car capable of autonomously navigating a moderately crowded freeway in the Bay Area.

Stability control systems and adaptive cruise control systems already show that car computers can make critical decisions.

But complex tasks such as merging onto a freeway or making left turns in traffic are significant challenges, Thrun admits.

If they ever do get on the road, such cars could transform society. Imagine a commute where you were free to work, read or perform other useful activities as your car drove you to work.

In a 2005 study, the Texas Transportation Institute estimated that American drivers on average lose 47 hours per year in congested traffic. The lost productive time costs the U.S. economy $63.1 billion annually .

No longer would parking lots have to be next to offices or schools, because cars could be programmed to park themselves elsewhere after dropping off their occupants. Such effects could transform urban real estate, insurance and the auto industry.

It sounds too good to be true, but Thrun is a believer. The biggest challenges are not technical but societal, he says.

Although robotic cars could save thousands of lives, they would not eliminate fatal accidents. The question is whether people would accept fatal robotic errors.

“There is a very significant legal problem,” Thrun acknowledges. “Whatever we build will kill people at some point.”

As it is, a lot of people don’t want to play chess with a computer or have a computer drive their car.

“None of us think about a world where all the cars are automated all the time,” Thrun said. “It could take society 20 years to adopt the technology.”

Ralph Vartabedian can be reached at