Wake-up call

At Carnegie Mellon, researchers are developing a warning system that would alert drowsy drivers when it's time to pull over.

By Peter A. Bradley -- Logistics Management, 6/1/1999

It's a feeling far too many drivers know. After long hours on the road, concentration becomes more difficult. Eyelids droop. Sleep beckons, regardless of the danger. Too often, drivers push on.What's potentially deadly for the driver of an automobile is much more dangerous for the driver of a 50,000-pound tractor-trailer. That's why the debate over hours-of-service regulations for truck drivers is intensifying. Safety advocates and motor carriers both want to ensure that only responsible, safe, well-rested, and alert drivers are on the road. As the accompanying story explains, the debate is over which rules can best assure that goal is reached.

Even a driver who takes the most stringent precautions may tire behind the wheel. Illness, a night of fitful sleep, the labor of loading or unloading a trailer, and other factors can drag the operator toward sleep. According to a Federal Highway Administration study, a significant percentage of heavy-vehicle accidents are fatigue-related. With that in mind, researchers at the Carnegie Mellon Robotics Institute have been developing an in-cab warning system that essentially would function as an alarm that tells a driver "You're too tired to be behind the wheel right now."

Virtual Road Test

In a converted warehouse in downtown Pittsburgh, Carnegie Mellon researchers are testing a system that uses an infrared light to measure how often drivers' eyelids droop, indicating they are drifting toward sleep. To test the system, they are using a sophisticated driving simulator that realistically mimics a wide variety of road conditions, including changes in congestion, weather, and time of day. Pittsburgh-based Pitt Ohio Express, a regional less-than-truckload carrier, is working closely with the Robotics Institute on the project, and 16 Pitt Ohio drivers have volunteered as test subjects. Pitt Ohio also has made trucks available to researchers so they can test the system as it evolves.

The simulator, developed by ISIM Corp. of Salt Lake City, Utah, is housed in a darkened room next to a bank of computers and video screens that are used to monitor the drivers' performance. The drivers sit in the cab of a Freightliner tractor. When the simulation begins, the highway comes to life. On screens that wrap around the front of the room, the driver sees the road ahead, traffic, and even the sky. Rearview mirrors show the scene behind. As the driver shifts and presses the accelerator, the tractor appears to move over the road. Turn the wheel, and the vehicle turns. Press the brake, and the driver can feel the load shift behind him.

The simulator is an invaluable tool for the researchers because it allows them to evaluate performance accurately without compromising driver safety in any way. "We are trying to find a non-invasive way to measure drowsiness that would alert drivers that they are in a state that may be dangerous," says Dr. Richard Grace, senior systems scientist at the Robotics Institute and head of the National Highway Traffic Safety Administration project.

The Carnegie Mellon researchers, together with colleagues at Virginia Technical Institute and the University of Pennsylvania Sleep Laboratory, sought an appropriate measure of drowsiness for nearly two years. They decided to use what they call the percentage of eye closure over time--or "perclos" in the researchers' shorthand. When people get tired, their eyelids begin to droop and close. But using drooping eyelids as a reliable measure of drowsiness presents several problems for the researchers. First, any such measure must distinguish normal blinking, which takes half a second or less, from the slow eye closures that indicate drowsiness. It also must account for individual differences. And any system for measuring eye closure must not physically interfere with or restrain the driver.

That's why the volunteer drivers are so important. The drivers come to the Robotics Center after their normal shift at Pitt Ohio. As part of the agreement, the Robotics Center has guaranteed that it will not divulge any individual performance details to the company or any agency.

During the testing phase, each driver spent several hours in the simulator. Mounted on the dashboard of the tractor was a small device that emitted a low-level infrared light. The light intensity was a small percentage of the maximum allowed by federal safety limits. In addition, cameras in the tractor focused on the driver's face from several angles. The infrared light was invisible to the driver, but researchers watching screens outside the simulator could see the driver's eyes. On a screen showing only infrared light, the driver's eyes appeared as two glowing red embers; this was caused by the reflection of infrared light from the eyes. When the driver blinked or closed his eyes, the reflections disappeared. Cameras and other specialized devices in the truck collected data on how often the driver closed his eyes and for how long.

Next, trained scorers examined the data and counted eye closures. After analyzing the results of countless hours of observations, they settled on 12-percent eye closure over the course of one minute as a reliable indicator of drowsiness.

The Importance of Accuracy

"The next step was to automate the system," says Grace. "No matter how good the scorers were, they were not as objective as we wanted.'' Then came development of a computer program that could find the driver's eyes, measure how often the eyes closed, and trigger an alarm that would alert the driver to pull over and get some rest.

The system has worked well in the test phases, but it still is a ways from becoming available to trucking companies or to the public. Grace hopes to do a large-scale field test with 100 or more trucks. One problem that still remains, he explains, is that the infrared light reflects well off the eyes of some people, but not so well in others. "We want to find an infrared that works with a broader range of people," he says.

To make the application practical, the researchers want to develop an in-cab system that is compact, self-contained, and affordable. The most critical issue in making the system practical, says Grace, will be gaining driver acceptance. That's why [ensuring that the] system is accurate and reliable before rolling it out is so important. "A lot depends on the drivers," he says. "If they feel it's accurate, they'll stop and take a nap. If they don't feel it's reliable, they won't."

Paramount among the researchers' goals is making sure any system does not inadvertently compromise safety in any way. "We're treating it as if we were developing a drug," Grace says. "We want to make sure we do no harm. We want to make sure drivers' reactions are safe."