The divided highway lances through the beauties of a Pennsylvania dawn, morning frost glinting across the hills below. It’s hard to enjoy the natural wonders, though, when ALVINN is behind the wheel, doing 88 km/h in the fast lane. First he lurches right, crossing both lanes of the blacktop and alarming bleary-eyed commuters trying to share the road. Then he careens to the left, skidding onto the gravel shoulder and nearly into a ditch. Finally Todd Jochem, 27, a graduate student at Pittsburgh’s Carnegie Mellon University, wrests the wheel of the four-wheel-drive Humvee from ALVINN while Dean Pomerleau, a C.M.U. robotics research scientist, makes excuses for their friend’s driving. “I guess he’s a little confused,” says Pomerleau. “We’ll let him try again in a minute.” Still, ALVINN is a marvel–a road-hogging computer able to drive like a human. Well, almost.
ALVINN, short for Autonomous Land Vehicle in the Neural Network, originated in 1985 as a research project by C.M.U.’s Robotics Institute for the U.S. military. Early progress aroused the U.S. Department of Transportation and U.S. automakers to the fascinating possibilities of fully automated passenger vehicles. The lean-back comfort of a train meets the door-to-door convenience of a car. Automobile travelers would be able to key in their destination, flip open a magazine and leave the actual driving to ALVINN. Last October the DOT committed $160 million to an Automated Highway System research consortium, which includes the C.M.U. team. A European coalition is working on similar technology called Prometheus; Japanese automakers are also tinkering with prototypes.
At the moment, ALVINN can go only where he has gone before. The technology– cameras, laser range finders, sonar and motors–works in three stages. First ALVINN “sees” the road by taping it as well as sending out laser and sonar waves as Jochem drives. Then the computer goes into training, replaying the tape thousands of times, studying the details–road signs, lane markers, speed limits. Finally ALVINN drives a stretch of road, handling his own acceleration, braking and steering.
Right now ALVINN is barely a student driver. Small but unexpected events cause him to swerve and shake in gigabyte panic. His performance may improve this month when “smart car” map scanning is integrated into the programming. “There should be a substantial difference,” promises Pomerleau. The researchers say they are only a couple of years from achieving a preliminary goal, perfecting ALVINN as an anticollision device to jog sleepy drivers before they run off the road. When a car drifts dangerously close to the border of its lane, a road-watching camera would do something like ring an alarm bell or shake the driver’s seat. “Fully automated driving may be a ways off,” says Pomerleau, “but in the near term we’re doing work to save lives.” For now, however, if you see ALVINN coming at you on the road, you might want to keep a respectful distance.
More Must-Reads from TIME
- L.A. Fires Show Reality of 1.5°C of Warming
- Home Losses From L.A. Fires Hasten ‘An Uninsurable Future’
- The Women Refusing to Participate in Trump’s Economy
- Bad Bunny On Heartbreak and New Album
- How to Dress Warmly for Cold Weather
- We’re Lucky to Have Been Alive in the Age of David Lynch
- The Motivational Trick That Makes You Exercise Harder
- Column: No One Won The War in Gaza
Contact us at letters@time.com