Before automating a part of the driving experience, you’d like to know how much it will cost and how much safer the passengers will be. The first problem is fairly easy for bean counters to solve, but the second one is a bear. It’s hard to predict how such an invention will work in the real world.
Take collision-avoidance and lane-departure warning systems. The systems are too new and too rare to have been in a lot of accidents, so you can’t just compare cars that have them with cars that don’t. Instead, you have to start with examples of the accidents you’re trying to prevent and deduce exactly what would have happened if these automated systems had been in place.
In a project supported by Toyota, engineers at Virginia Tech have worked out a particularly detailed way to do this. H. Clay Gabler, Kristofer D. Kusano, and Thomas I. Gorman started with representative crashes. They then built a computer model to play through each case, reconstructing the initial conditions and simulating what would have happened if the driver had been warned.
In one experiment they took a sample of 2484 accidents due to lane-departure errors from some 740 000 such accidents—the total recorded in the United States over a four-year period. The accident reports had only some of the data needed to simulate the possible trajectories of the cars. So, to fill in the blanks, the researchers used statistics on speed limits, shoulder width, radius of curvature of the road, and so on.
They inferred the speed of the vehicle from the extent of the damage in the crash. They then ran two simulations, one assuming that the driver would take just 0.38 second to react to a warning, the other assuming he'd need 1.36 seconds. The engineers also assumed that the longer the car moved along an off-course trajectory, the greater the chance of a collision would be.
Their conclusion: lane-departure warning systems would have prevented 30 percent of all the accidents due to a mistaken departure from the lane.
In another experiment, the researchers worked instead with data collected on a test track, from cars using 10 different lane-departure warning systems. They calculated that various systems could prevent between 29 and 32 percent of crashes.
A similar study of 10 forward collision warning systems found far greater differences, preventing as few as 9 percent and as many as 53 percent of rear-end collisions. It all depended on the speed at which the systems kicked into action. The researchers concluded that it would be a great idea to use these rear-end prevention systems not only at highway speeds but also below 40 kilometers per hour (25 miles per hour).
You may well ask why anyone bothers to model the advantages of these warning systems when it stands to reason that they must save lives. Problem is, what stands to reason hasn’t always turned out to be true.
The early antilock braking systems (ABS) seemed so obviously good that the public flocked to buy them as optional features, and insurers offered discounts on policies for cars equipped with them. But when the accident reports rolled in, insurers found that ABS had made no visible improvement. It seems that drivers, emboldened by their super-automatic brakes, had driven a little more aggressively than before.
Of course, if the gains of automation are drastic enough, then even drivers' riskier behavior can’t eliminate them. Later ABS systems that incorporated vehicle stability functions did save lives. And if you throw a bodyguard of robotic systems around your car—lane-departure warning; front-collision warning; driver drowsiness warning; cross-traffic collision warning—eventually it will be positively hard to use a car to hurt yourself or others.
Philip E. Ross is a senior editor at IEEE Spectrum. His interests include transportation, energy storage, AI, and the economic aspects of technology. He has a master's degree in international affairs from Columbia University and another, in journalism, from the University of Michigan.