All Tech Considered
Mon August 19, 2013
Hitting The Road Without A Driver
Originally published on Mon August 19, 2013 1:19 pm
The cars we drive have gotten ever more sophisticated. They can just about park themselves; they tell us if we're drifting out of our lane; they can prevent skids. Some even automatically apply the brakes if they sense that a collision is imminent.
Engineers at Carnegie Mellon University in Pittsburgh are developing a car that can do all of those things and more — it can actually drive itself. Imagine that commute to work.
The car, developed with General Motors, is by all appearances a normal Cadillac SRX crossover. That's by design, according to Jarrod Snider, the chief engineer on the project.
"We're not using really fancy, really expensive devices all over the car," he says. The sensors are integrated into the car, "so when you look at the car you don't see a lot of things hanging off of it."
Hidden in the bumper and behind the car's grille are three types of systems that guide the car's decision-making — sensors, lasers and cameras. Snider says the sensors provide input for the car's software, "so when we see an object we can say that's a person or that's a sign or that's a traffic light. ... We can actually do some classification of the object."
The car's navigation screen will show and identify pedestrians and bikes, as well as traffic lights and some road signs. It also has thermal imaging, useful to identify objects at night, beyond the range of the car's headlights.
In the back of the car, below the floor where the spare tire would normally be, are four computers about the size of Apple's Mac Mini, cooled by an air-conditioning system. Snider says the computers are a development platform and won't be as big or noisy when driverless cars are commercially available.
Inside, the car looks pretty normal, too, with one exception — a big red button that sits prominently in the middle of the dashboard. Snider says it's there for when engineers are testing a new system. If something goes wrong, "we can just push that button and that just turns the car back into a stock car, immediately," he says.
Surprisingly, a test ride in the autonomous Cadillac occurs not on a closed course, but on busy Route 19, a multilane highway in Cranberry Township, outside Pittsburgh. Snider says township officials have been cooperative and helpful with development of the driverless car, installing special radio transmitters on traffic lights to tell the vehicle when to stop (the car can also "read" the lights with a camera positioned over the windshield).
Snider says it's hard to replicate real-world conditions on a closed course. People drive in unpredictable ways; they stop suddenly, change directions; there are pedestrians jaywalking.
To put the car in driverless or autonomous mode, Snider simply turns a knob located near the gear shifter.
"System starting up," a female-sounding computer voice intones. "Autonomous driving."
The car pulls out of the parking lot. Though Snider is behind the wheel, he's not touching it, as the car stops at a stop sign, waits for traffic to clear and then makes a right turn onto the highway.
The car is programmed much like a driver programs a GPS for directions. In this case it's heading for a restaurant in a shopping center. We travel about a half-mile down the busy road, the car accelerating, stopping at a red light and moving into a left-hand turn lane, where it waits for a green arrow before proceeding.
The drive is not always smooth. The car tends to wait until the last minute before braking and floors it when accelerating to its desired speed. Snider says engineers are working on that, but that it's difficult to precisely re-create the human touch on the gas pedal and brakes.
Carnegie Mellon's car is not the only driverless vehicle around. Google has one as well. Raj Rajkumar, a professor of electrical and computer engineering at Carnegie Mellon, predicts that by 2020 the technology needed for driverless cars to travel normal roads will be ready. Though he says it make take a bit longer for the legal system and insurance companies to catch up.
"It has to go through the societal process of acceptance, the legal process of laws being in place that allow driverless vehicles on the roads, the insurance aspects of liability, the legal things falling into place," Rajkumar says.
The federal government has already begun conducting research with an eye toward establishing standards for driverless vehicles. Three states — California, Nevada and Florida — have passed legislation allowing testing of driverless cars on their roads.
RENE MONTAGNE, HOST:
Let's head to the streets, where the cars we drive have gotten ever more sophisticated. They can just about park themselves, tell us if we're drifting out of our lane and prevent skids. Now we may have arrived at the final frontier, putting all those things together and more to make cars that drive themselves. The latest prototype has been produced by engineers at Pittsburgh's Carnegie Mellon University. NPR's Brian Naylor took a spin.
BRIAN NAYLOR, BYLINE: The first think you notice about Carnegie Mellon's autonomous car is, well, nothing. Sitting in a parking lot, it looks like any other medium-sized SUV. Jarrod Snider is the project's lead engineer.
JARROD SNIDER: This is a 2011 Cadillac SRX, of course heavily modified by us. Most of the sensors are integrated in, you know, the bumpers or somewhere in the car.
NAYLOR: Hidden in the bumper and behind the car's grille are three types of systems that guide the car's decision making - sensors, lasers and cameras.
SNIDER: So when we see an object we can say that's a person or that's a sign or that's a traffic light, something like that. We can actually do some classification of the object.
NAYLOR: We walk around to the back of the car. Snider opens the hatchback and lifts the floor where the spare tire would be.
SNIDER: So this is where the main computing happens. Inside here we have four computers. They're all right now doing different tasks, but they're essentially identical hardware.
NAYLOR: The computers are small black boxes about the size of dinner plates. The sound is the air conditioning to keep them cool. They won't be this big and noisy when driverless cars are commercially available. And inside the car looks pretty normal too, with one exception. There's a big red button that sits prominently in the middle of the dashboard.
SNIDER: That's an emergency stop button. That's basically when we have experimental development code, we're testing something new. If something goes wrong, we can just push that button and that just turns the car back into a stock car immediately.
NAYLOR: Somewhat reassured, I climb into the back seat. Snider takes the wheel and Carnegie Mellon Professor Raj Rajkumar rides shotgun as the car's inner voice speaks up.
COMPUTER VOICE: System starting up.
SNIDER: It tells me that it's starting up. An autonomous ready means that we can now put into auto whenever we want. So it's in manual mode. I can drive like a normal car. I mean it's letting me know that whenever I want the car to take over, it can take over. So I'll go ahead and put into autonomous mode here.
VOICE: Autonomous driving.
NAYLOR: The car is programmed much like a driver programs a GPS for directions to, say, a restaurant. I assume it's programmed to take us on a spin around a closed test course, but no, that's not where we're going at all.
SNIDER: And basically what we've done is we've asked it to take us to a complex down here that has some restaurants and things like that there. So this is simulating that you start at your house and you want it to take you to dinner, something like that.
NAYLOR: And I should point out, we're not on a little country lane here. We're on a four-lane, actually a five-lane divided highway in the middle of a busy suburb with traffic lights, with shopping centers and restaurants and quite a lot of traffic. And this car is making the decisions on when to accelerate, when to stop, when to change lanes, all on its own.
Snider says it's hard to replicate real world conditions on a closed course. People drive in unpredictable ways. They stop suddenly, change directions. There are pedestrians jay-walking.
SNIDER: Given that people can do anything like that, we have to make sure our car is pretty robust and pretty safe and can deal with things that aren't just, you know, the nominal driving that people do.
NAYLOR: Carnegie Mellon's autonomous car can detect red lights. It recognizes pedestrians and bicyclists. Still, there are a few glitches to be worked out.
VOICE: Changing lane.
SNIDER: Sorry. There's a sign up there. It wants to change lanes, but there's a car there that's not allowing it to change lanes. Even though the turn signal is on there (unintelligible) so I think we'll wait for an opportunity and then it will - and then once the traffic is clear...
VOICE: Changing lane.
NAYLOR: The car also tends to wait until the last minute to brake and floors it when accelerating, but Snider says those things are being worked on. Carnegie Mellon's car was developed in a partnership with GM and it's not the only driverless car around. Google has one as well. Carnegie Mellon Professor Rajkumar predicts that by 2020, seven years from now, the technology needed for driverless cars to travel normal roads will be ready.
Though he says it make take a bit longer for the legal system and insurance companies to catch up.
RAJ RAJKUMAR: It has to go through the societal process of acceptance, the legal process of laws being in place that allow driverless vehicles on the roads, the insurance aspects of liability, the legal things falling into place.
NAYLOR: The federal government has already begun conducting research with an eye toward establishing standards for driverless vehicles. And three states - California, Nevada and Florida - have passed legislation allowing testing of driverless cars on their roads. Brian Naylor, NPR News. Transcript provided by NPR, Copyright NPR.