Carnegie Mellon developing driverless car of the future now
Goal is to replace error-prone humans
August 25, 2013 8:00 AM
Mounted on the dashboard of the computer-driven Cadillac is a thermal infrared camera used to help the car categorize the objects in its path.
Jarrod Snider monitors the computer-driven 2011 Cadillac SRX as it makes its way Thursday on Route 19 in Cranberry.
By Michael A. Fuoco / Pittsburgh Post-Gazette
Route 19 north in Cranberry is crowded with three lanes of cars zooming during Tuesday's lunch hour. A metallic gold 2011 Cadillac SRX with four passengers is on the clogged highway, smoothly traveling at the 45 mph speed limit. Suddenly, danger looms -- a black Jeep slowly enters the roadway right in front of the Cadillac. The Cadillac quickly brakes without losing control, slows down and continues moving, avoiding crashing into the rear of the erratically driven Jeep.
Now that was a pretty nifty piece of driving -- especially by a computer.
Yes, that's right, the Cadillac was totally, completely driving itself (with onboard human monitoring, to be sure). It stopped at stop signs and red lights, safely entered traffic on cross streets, made turns (using turn signals, natch), changed lanes to pass cars, and slowed down and accelerated at appropriate times, all the while choosing the most efficient route to a pre-programmed destination.
Autonomous driving is hands-off experience
Enigneers from the General Motors-Carnegie Mellon Autonomous Driving Collaborative Research Lab at CMU demonstrate their autonomous driving 2011 Cadillac SRX along Route 19 in Cranberry. (Video by Rebecca Droke; 8/25/2013)
Looking like any other Cadillac SRX, the crossover is actually a high-tech experimental self-driving vehicle under development at the General Motors-Carnegie Mellon Autonomous Driving Collaborative Research Lab at CMU in Oakland. The goal of the project's 15-member team -- CMU faculty, full-time research staff, post-doctoral researchers, doctoral students and master's students -- is to develop an autonomous driving vehicle that can more safely navigate highways than humans can.
Each year, more than 30,000 people die in vehicular crashes in the United States. Another 2.3 million hurt in crashes require emergency room treatment. Since 90 percent of all accidents are caused by driver error, improving vehicular safety is the main reason GM/CMU, Google and other scientific entities are in the process of developing cars that some day will drive us. All we'll have to do is tell it where to go.
"Humans are extremely smart but can be rather stupid as well," said CMU professor Raj Rajkumar, co-director of the autonomous driving research lab. "We can be distracted, angry, mad, sleepy, drunk, driving while eating, shaving, applying makeup and even changing clothes. If we can take the basic human emotional and physical problems out of the equation, we expect injuries and fatalities will go down."
Mr. Rajkumar, who earned his master's and doctorate in electrical and computer engineering at CMU, said autonomous driving vehicles also will make passengers more productive during the average work week they spend sitting in stalled traffic each year, enabling them to catch up on email, reading or even napping. And it will provide mobility for elderly and disabled people.
"What used to be science fiction is certainly no longer science fiction," Mr. Rajkumar said.
Indeed, while it may not be the flying cars that "The Jetsons" promised wide-eyed baby boomers, the concept of self-driving vehicles no longer is fiction. But there's certainly a lot of science involved -- computer engineering, computer science, mechanical engineering and robotics.
Here, then, is how all of those specialities fuse for a ride in the future.
• • •
A car that can drive itself might be expected to have science-y gizmos all over it, but the Cadillac parked in Cranberry Municipal Park's lot sure doesn't stand out. That's key for conducting real-world experiments because other drivers might alter their behavior otherwise, said Jarrod Snider, the project's lead engineer, who like Mr. Rajkumar and six other team members is wearing a blue polo shirt bearing "GM LAB -- Autonomous Car -- Carnegie Mellon."
"No one's aware of how much they're helping us out," he said, climbing into the "driver's seat" with Mr. Rajkumar sitting in the front passenger seat. "One of the big benefits of us testing here on public roads is that it is impossible for us to come up with this stuff on a test course. We have the randomness of the road and that's a good way for us to test."
The vehicle is equipped with six lasers and six radar covering all 360 degrees surrounding the vehicle and cameras in the front and back, including a thermal infrared model. Fusing all of the information gleaned from those devices are four whirling computers stored in a compartment beneath the cargo area that is easily accessed by lifting a panel.
Operating with 500,000 lines of computer code and in some cases making calculations in 10 milliseconds, the computers safely operate the vehicle's speed and direction of travel while determining lane markings, speed limits, the status of traffic lights, the speed and location of other vehicles, and classifying objects as humans, bicycles, motor vehicles or non-mobile obstacles.
Inside, it's hard to miss a red button in the middle of the dash reading "Emergency Stop." On the console is another knob, which can switch the car back and forth from manual to autonomous operation. In the middle of the dash and in the headrests in the back seats are video monitors that in different modes -- live cameras, animation and thermal infrared, among others -- illustrate the location of the Cadillac in relation to everything around it.
Mr. Snider touches a button on the display screen.
"System starting up. Autonomous ready," says a dulcet female voice.
He turns the console knob.
"Autonomous driving," the voice says and the car automatically shifts into "Drive" and begins moving. For a brief second, it's impossible not to think of HAL 9000, the male-voiced spacecraft computer in "2001: A Space Odyssey." Unlike HAL, this computer doesn't have a malevolent mind of its own but it is nonetheless surreal. The car accelerates, the steering wheel turns and the vehicle pulls to a stop at a stop sign -- all while Mr. Snider does nothing but watch.
From all the information being gathered by its myriad sensors, the vehicle knows the road is clear and it continues down Ernie Mashuda Drive, clicks on its right turn signal and stops at the stop sign at the intersection with Route 19. Cars whiz by and the Cadillac waits and waits. And then, when it is safe to pull out, it turns right and begins moving at 45 mph, the speed limit.
"Changing lanes. Changing lanes," the female voice says, and the car does just that.
On its monitor, the vehicle shows it is in the middle lane with vehicles in front, on either side and behind it. "Caution. Entering Construction Zone," the voice announces after "reading" a sign alongside the roadway.
The car approaches traffic lights as they turn red. The car stops at the intersection. The computer has a bit of a heavy foot on the brake, like a 16-year-old learning to drive.
"It doesn't do anything that's unsafe, but it doesn't have the finesse a human has," Mr. Snider notes. "A human will really kind of round off the stop so it's not abrupt. Those kinds of things we look at also: How can we model those kinds of complex behaviors?"
The light turns green and the Cadillac moves along with the traffic. The surrounding drivers have no idea they are riding alongside the future.
After about two miles, the computer announces it is changing lanes, puts on its turn signal, eases into a turning lane and makes a left, crosses Route 19 north and then makes a quick right into a strip mall at 10 Francis Way. That was the destination Mr. Snider had given the computer. He then touches a button on the screen and the car drives itself back to the municipal park.
There, while stopped, it identifies on the screen pedestrians walking in front of it. Once it's safe, it begins to drive again, pulls behind a bicyclist, identifies it as "Bike" and slows to a safe distance behind it.
For the vehicle, which has yet to be nicknamed, it had been another successful drive at moderate speed in a traffic-controlled environment. Up next, testing on high-speed highways.
• • •
Mr. Rajkumar estimates it will be decades before totally autonomous driving will be commonplace on the country's roadways, but as his team and other researchers perfect new autonomous functions, car makers will incrementally offer them in consumer vehicles. For example, features such as on-demand autonomous parallel parking, lane-drifting assist and adaptive cruise control, among others, are already available.
"Computers in cars have been doing some of the controls for quite some time and they will be doing more and more," he said. "To me, the next coming generation -- the next 30 years or so -- will see this hybrid, where advance technologies do more and more driving but coexist [with a human driver]."
But, he added, widespread use of autonomous driving vehicles is inevitable.
"This evolution, when it happens -- not if, but when -- will be very gradual, very incremental," he said.
"But I do see a point in time [when the feeling will be] that if you are human, you are capable of making mistakes, and therefore if you are driving, you could be a danger to yourself or others. And because of that, you definitely would not be able to drive on public roads."
Mr. Rajkumar got in his personal car and pulled away, his hands on the steering wheel, his foot on the accelerator, his eyes on the road.