Can a robot car survive city traffic?

Industry news

Washington DC - The car is tootling around the US capitol - pretty much on its own - when a police officer jumps into the road ahead of it, almost within spitting distance of the Capitol dome.

What is the cop waving about? Hard to say. The car is being driven by computers, and wild waving is a bit too complicated for them to understand.

Tell a friend
Carnegie Mellon University Professor Raj Rajkumar (L) invites U.S. Acting Transportation Undersecretary Peter Rogoff (R) for a drive in the university's autonomous car in Washington June 24, 2014. The 2011 Cadillac SRX is able to control steering, speed and braking in harmony with the autonomous systems (radar, lidar and infrared sensors) that detect and avoid obstacles in the road.  REUTERS/Gary Cameron   (UNITED STATES - Tags: POLITICS TRANSPORT SCIENCE TECHNOLOGY BUSINESS) - RTR3VIBSCarnegie Mellon University Professor Raj Rajkumar is shown with the main, multi-computer control center in the rear cargo area of a self-drivien car that was on display in Washington on June 24. The car has been retrofitted with cameras and sensors to detect traffic lights, pedestrians and read some road signs. Illustrates DRIVERLESS-CAR (category l), by Ashley Halsey III © 2014, The Washington Post. Moved Tuesday, Aug. 26, 2014. (MUST CREDIT: Washington Post photo by Linda Davidson)

Passenger Jarrod Snider taps a button on the center console and puts his hands on the steering wheel.

“Autonomous ready,” the voice of the computer says a fraction of a second later, eager to take control again.

Swing a stick on the Mall this summer and you'll hit a dozen sceptics who doubt that the streets of Washington - or any city - will ever be filled with cars that drive themselves.

But the doubters may well witness that transformation in their lifetimes, and very likely sooner than they think.

The ability of the vehicle cruising unnoticed among the tourists and important people in pinstripes on Capitol Hill would shock most of them. A ride in it also points to a few chinks in its armor.

The computers running the car, for example, can see the police officer bustling into the middle of Constitution Avenue. But they can't figure out why he’s doing it - and neither can the people riding in the car.

It turns out the officer wants to pull over a driver in another car who was making an illegal turn.

Could the car have handled it without Snider's help?

“Yeah, it started to slow down before I took over,” Snider says, “and as he stepped out of our lane and walked across the street, the car would have continued to go. The car obviously doesn't understand gestures like 'Stop here.' “

If this car - a silver-gray Cadillac SUV converted to autonomous driving by Carnegie Mellon University - looked the least bit odd, the Capitol Police would swarm after it with machine guns.

It doesn't. But it's bristling with technological weapons.

Two cameras - one pointing up at traffic signals, the other down at lane lines - are hidden beneath a slight ridge added just above the windshield. There’s long-range radar behind the Cadillac medallion on the grille and short-range radar - as well as a pair of laser beams - behind the front bumper.

Unseen behind tinted windows near the back seat, from unobtrusive boxes that match the Cadillac's tan interior, a radar and a laser beam look out to each side. From the rear bumper, more radar and lasers.


All of them feed into a bank of four computers hidden in the spare-wheel well beneath the rear floor of the vehicle, along with GPS data and mapping feeds. They know speed limits and, unlike the other driver on Constitution Avenue, places where left turns are illegal and where right turns on red are allowed. If one computer fails, the others take over its tasks and the person behind the wheel gets an alert.

Right now, put the Cadillac on an interstate and its developers say it could drive you from Washington to San Francisco, although it would need your assistance at fuel stops.

But in Washington? That's another matter.

The city's streets are full of cars driven by impatient locals and bewildered tourists. Pedestrians talking on cellphones, texting tourists, cabs darting across lanes to grab a fare, cyclists by the dozens and long-distance tour buses whose drivers appear to be feeling their way around town.

“We've tested this vehicle in many places, and this area is pretty difficult,” Snider says, dropping his hands from the wheel as the computers take over again. “We think it's doing a pretty good job.”

“But we have more work to do,” Raj Rajkumar, the Carnegie Mellon professor who directs the project, says from the back seat.

The car buzzes on down Constitution Avenue, switches on its right indicator, slows and then turns south on First Street. It identifies a red traffic light and dutifully stops in front of the Supreme Court until the light changes.

Then there's trouble.

A white delivery truck is stopped in the right lane just past the light, and a yellow cone behind it says it isn't going anywhere soon.

“I'm going to take over to get us past,” Snider says, punching the console button. “It doesn't have the higher-level reasoning like we have that there's a cone there so this truck's probably not going to move - so it'll just sit there.”

The computers are back in command as the car nears Independence Avenue, signals its intent to turn right and then stops to wait for the red light to change. This traffic light is one of six in the District that alert the Cadillac to its color. Someday all lights may do that, but the car's cameras don't really need the help.

The intersection is swarming with the lunch-hour crowd, and each pedestrian in the crosswalk or sidewalk shows up as a green squiggle on the standard dashboard screen that Cadillac builds into all its vehicles. In this one, however, the screen displays 360 degrees around the car: people, obstacles, traffic signals, construction zones and other vehicles.

The big red bus in the right lane on Independence Avenue is the Cadillac's next challenge.

“Changing lanes,” the computer announces, moving to get around it - but the bus driver apparently doesn't like that.

“We're trying to pass him, but then he cut us off,” Snider says.

If she is upset - and the firm but melodious voice of the vehicle unquestionably belongs to a woman - she doesn't let on. Her verbal skills are limited; in addition to “Autonomous ready” and “Changing lanes,” she says “Starting up,” “Entering work zone” and “Exiting work zone.”

The rest of the computers' communication - currently and what's planned in the future - come in chimes, beeps and vibrations.

If the person in the driver's seat touches the wheel or either of the floor pedals, much as with cruise control, the computer relinquishes control. If the computer needs the driver to take over, the steering wheel and passenger seat may vibrate.

“Sometimes, if it’s not confident about something, it can tell you to take over, and if everything's OK, it can tell you it's ready to drive autonomously,” says Snider, lead engineer on the project at Carnegie Mellon. “It's just providing some feedback to the driver.”

Then it's time to make a tricky left turn from Independence Avenue onto Washington Avenue, known as a difficult intersection.

“It doesn't have a green arrow here, so obviously it's got to detect the cars coming from the other direction,” Snider says as the car waits patiently, then makes the turn, veers right on to Second Street and then takes the on-ramp to Interstate 395 south.

As two merge lanes that lead to the highway come together, a silver Mercedes suddenly forges ahead and Snider pops the console control button.

“Maybe I should have let the computers do that,” he says with regret after the merge is complete. “It definitely has some trouble with that because of the drivers here. The car doesn't have that kind of aggressiveness. It won't push its way in or force its way into a merge.”

Back in control, the computers accelerate the car to 80km/h as it crosses the bridge into Virginia, surrounded by other cars.

“It's determining how fast to drive based on the curvature of the road, based on the other cars in front of it,” Snider says.

“It knows the speed limit.”

It's not going to break the speed limit, but it can obviously do other things, like here you can see it's slowing down to make its way into the other cars.”

“Changing lane,” the computer chimes in.

“All of this is being updated in real time, determining what to do,” he says. “So now it's just taking an exit here.”

The car zips off I-395, takes a right in front of the Pentagon and then neatly merges back onto I-395 headed north. Over the bridge, it will peel off the freeway back into Washington traffic and complete an uneventful trip back to a parking space beside the Capitol reflecting pool.

Driverless cars are coming to the United States and rest of the globe, Rajkumar is saying as the Cadillac covers the final blocks.

“Absolutely no doubt at all,” he says, before quickly acknowledging the doubters. “I welcome their skepticism.”

“Technology cannot be stopped. We just have to make sure that it is safe, affordable and legal.”

Don't expect an autonomous car to land in your driveway with a “big bang,” he says.

Remember anti-lock brakes? How about cruise control? Those were the first steps, he says.

The next one coming in production cars - within three to five years - will be a highway pilot feature, he says. Put the car in the correct lane, tell it to go to San Francisco, and it will.

A year or two later, highway “plus-plus” will arrive, allowing that San Francisco-bound car to weave around the slowpokes along the way.

In the same time frame - three to four years - look for traffic-jam assist capability. The car will take over for you while inching through bumper-to-bumper traffic and alert you to take back control once there's clear sailing.

“The totally driverless version will happen in the 2020s.”

But Rajkumar adds: “The whole process will be incremental. More and more scenarios that we drive in will become automated, until one fine day you've given up complete control without even noticing.”

Tell a friend