Cars Are Learning to Drive

Grandma has to see her cardiologist but you suddenly find you’re not available to take her. She’s willing to drive but even with her glasses, she no longer could pass any state’s eye exam. What to do? Just program the car to take her to the clinic, and then home again.

Uh oh, soccer-dad. It’s your turn to carpool but the kitchen sink is leaking and you don’t trust towels to sop up the deluge for even a 20-minute absence. Luckily you can program your van to pick up each kid, drop them all off at the soccer field, and then reverse the five-stop commute 90 minutes later.

These are the fanciful scenarios Raj Rajkumar of Carnegie Mellon University outlined for me when I visited his Pittsburgh office three weeks ago. I wanted to learn more about what General Motors expected his team to do for the $1 million it announced it would be giving these embedded-systems engineers for each of the next five years or more.

Frankly, the goals Rajkumar outlined sounded like a stretch. I wouldn’t trust grandma to make it into the correct doctor’s office even if she was dropped off at the curb outside the clinic. And rambunctious kids left alone in a vehicle navigating through traffic: That’s a prescription for disaster. I’d fear for the car, its electronics, and the sanity of everyone in adjacent vehicles.

But then Rajkumar painted a picture that I could envision. Let the car chauffeur a licensed driver who remains behind the wheel — to take over if the vehicle’s computer, its sensors, or some unforeseen calamity strikes. If all goes well, the driver/rider can sit back and read the paper during a congested morning commute, or a student might bone up on details for that test in 90 minutes.

This concept car, truth be told, sounds very appealing.

I’m one of those drivers who becomes drowsy as soon as I hit the interstates. That can make a solo trip to pick up my kid from college — a four-plus hour drive from home — a dicey proposition. But Rajkumar says that the system he’s working on would one day allow me to plug the coordinates for junior’s dorm into a car’s GPS system and then let me, at will, turn over primary responsibility for driving to my vehicle. If I nod off, that should be okay; the car’s on-board navigation system remains ever vigilant. And if I choose to drive, I can instruct the on-board commuter system to back off and let me zoom-zoom along Pennsylvania’s winding roads.

In theory, anyway, these cars should react faster to weather conditions and potential crises than people do, remain perpetually patient, and might even be able to communicate with any other robotic kin on the roads or in traffic-control centers to select the quickest route to a destination, based on ever-changing ambient conditions.

As you might well suspect, this vehicle exists only in the imaginations of automotive and systems engineers. However, Rajkumar’s team developed a proof-of-principle car — “Boss” — which last November won a race sponsored by the Defense Advanced Research Projects Agency. Shortly after that win, GM came to CMU and offered to finance the school’s development of an Autonomous Driving Collaborative Research Lab. Its creation, and Rajkumar’s directorship of it, was announced on June 19.

The DARPA race had pitted teams — some created by universities, others representing commercial groups — to build a car that could be programmed to drive along a somewhat-fixed course at a reasonable clip. But the robot car wouldn’t be allowed to violate rules of the road (such as going through red lights, failing to halt at stop signs, or driving in a lane against the designated flow of traffic). It would also be disqualified for causing an accident or failing to successfully navigate unforeseen obstacles (such as a truck that broke down in its lane, perhaps people trying to cross the road, or other vehicles that were driving erratically). Mixed in with the robot cars were vehicles driven by brave people.

In fact, no one was harmed during this experiment. It was staged at a former Air Force Base in a California desert that only simulated suburbia. Which was smart, since Rajkumar says these robot cars triggered “a good number of accidents” despite tooling along at a maximum of only 30 miles per hour.

Keep in mind, these were not grown-up versions of radio-controlled cars. Designers were allowed no real-time communication with their cars during the race. They couldn’t even watch them from the sidelines. Closed-circuit cameras allowed some delayed video of the contestants as they maneuvered toward the finish line.

Of 89 cars that tried out, just 36 qualified for the competition. Only six of these completed the 60-mile trek. And although the $2 million purse for first place may sound rich, Rajkumar cautions that their vehicular champ’s price tag “much exceeded that.”

But the real win for CMU, he said, was the proffered GM partnership to develop the next generation of sensors and computers to bring Boss’s grandchildren onto the market. For now, he jokes, at CMU “we’re the brains and GM is the brawn.” Indeed, he said GM hopes to scale up and test prototypes.

Developing ones that prove affordable and reliable are engineering issues that Rajkumar expects will be quite doable. The biggest hurdle to putting these cars on the open roads may, in fact, prove to be a social one: convincing insurance companies of the cars’ reliability so that they won’t be too skittish to indemnify robot chauffeurs.