Test driverless car on track, not street

Editorial

Experimental, driverless cars would be invited to operate legally on Florida roads soon if a flawed bill proposed by Rep. Jeff Brandes of St. Petersburg becomes law.

The goal is to send a welcoming message to high-tech companies with the potential to build unmanned vehicles. Come to Florida and experiment on our highways.

This bill is potentially hazardous, even though it requires robot cars to be monitored by humans who can quickly take control. The only reason to test a robot car in the chaos of everyday traffic is to find out what might go wrong. The right places to optimize safety are the test track, parking lot, cow pasture, anywhere but on busy public roads.

Google already is testing driverless cars with human monitors, so the topic is timely. Last year after Nevada authorized the testing of self-guided cars on its roads, Singularityhub.com gushed that the people of Nevada "stand to reap huge benefits from a transportation system that is potentially safer and more energy-efficient than what's on the streets today."

Similar wishful thinking seems to be behind the Florida proposal. But just because you can test a car in Nevada or Florida doesn't mean you would build it or its components here. Kitty Hawk did not become the center of the aviation industry.

Part of the bill is farsighted and correct. It gives the Department of Highway Safety and Motor Vehicles two years to report to the Legislature on suggested standards for licensing and operating a car without a driver.

That's enough time to explore the challenges, including liability for accidents and mistakes. Until then, the bill should be amended to allow experimentation on public streets only for brief demonstration periods under special permission and oversight. The bill would allow unlimited testing before the manufacturer, or possibly backyard inventor, even certifies the car meets safety requirements.

"It's hard to simulate a million miles of road driving," Brandes tells us in defense of his bill. He says engineers need real-world experience, such as what the car does if it encounters an elderly person walking across the road.

"The only way we learn," he says, "is let it ride."

We ask, what's the rush? It will be many years, if ever, before driverless vehicles become practical for consumers, taxi companies and trucking lines. First will come crash-avoidance systems, likely based on satellite-positioning, radar and vehicle-to-vehicle communication.

Your future car's computer might sense that another car is approaching a red light so fast it can't possibly stop without hitting you. A warning buzzer might alert you to slow down or stop. In some visions of this technology, the computer would brake for you.

Such aids, including self-steered parking now offered in some cars, fall far short of the technology this bill addresses. It foresees an artificial intelligence smart enough to choose to hit the dog on the sidewalk instead of the child who darted into the road. It expects the robot brain to be superior to the human to allow roads to carry more and faster traffic while eliminating crashes.

The military too is interested. Any foolproof machine that could navigate through traffic, brake for squirrels and understand the significance of a stopped school bus would be a formidable combat vehicle, useful at any price.

The Army Research Laboratory has been working on the challenges. Research is being done at appropriate places like the Aberdeen Proving Ground in Maryland.

Those are safe sites to find out what happens if a bug smashes into and obscures a sensor; if the computer or GPS blacks out at high speed; to find out whether the computer's IQ falls after being left out in the hot sun all day; or how it reacts should the accelerator stick. Let the machine prove it can see the difference between a mannequin and shrub. Experiments also will suggest what level of education is needed by the owners and how often the computerized controls need testing and adjustment.

Perhaps it would be instructive to race robot cars on a NASCAR track.

No, says Brandes: "We know they can go in a circle." But we don't know whether they can truly drive.

To turn them loose on today's roads would pointlessly put the cart in front of the horse.