”Robot Cars Drive Themselves!” Pretty grabby headline, right? Throw in a few million dollars, a lot of publicity, and you've got a great story. The TV footage is compelling: brightly colored vehicles without drivers, bristling with cameras and sensors, driving themselves over dunes, down rutted trails--and, late this month, through simulated suburbs and city centers, in the US $2 million DARPA Urban Challenge.

There's just one problem with the imagery: the technologies likely to win the Challenge--those expensive cameras and sensors--probably won't be the ones that let future passenger vehicles ”drive themselves.” Instead, automakers expect that cars of the future will pay less attention to the lay of the land and more to each other, informing other vehicles about what they're doing several times each second by transmitting data that cars today already gather, via cheap wireless transponders roughly equivalent to your US $40 Wi-Fi router.

DARPA's interests are not in replacing commuters but in providing new and better technology for waging war. The appeal of an autonomous tank or rocket launcher is obvious: without soldiers inside, the potential casualties are reduced to zero. And the Department of Defense is under a 2015 deadline for making 30 percent of the U.S. military's land vehicles autonomous.

The challenge is substantial. An autonomous military vehicle must negotiate every kind of terrain: sandy desert, muddy forest, and dense urban core. To a tank, everything is a potentially hostile obstacle. Aside from its own location, tracked via the Global Positioning System, it has to figure out where everything in its surroundings is in real time.

Passenger cars, on the other hand, operate in far more limited circumstances: they stay on roads, almost all paved. They have no need to hide themselves, operate stealthily, or attack other objects. (In fact, making themselves known leads to avoidance, and hence safety.) And there are 250 million vehicles in the United States alone, according to the U.S. Bureau of Transportation Statistics, so traveling at high speeds among many adjacent moving objects with constantly changing trajectories is crucial.

Modern cars are stuffed with microprocessors and electronic control units that process data from a huge variety of sensors in the engine, transmission, suspension, and other systems--and then deliver the right blend of performance, fuel economy, and safety. Already, many traction-control systems simply ignore what drivers ask the car to do if the actions would cause the car to skid. Their sensors, though, are limited to the mechanical phenomena the car itself is experiencing.

Several safety systems have now added environmental data to the mix. Adaptive cruise control, from Mercedes-Benz and others, is one. It uses radar to calculate the distance to the car ahead and that car's velocity and adjusts its own speed to maintain a safe distance at all times--braking right down to a standstill and then accelerating back to highway speeds.

Another is the Volvo Blind-spot Information System (BLIS), which scans the area around a car's rear corners with side-mounted cameras and alerts the driver if there's a vehicle present. A third is Infiniti's Lane Departure Warning system: it calculates the edges of the lane from images captured by a video camera behind the windshield and alerts the driver if the vehicle is about to drift too far.

All of these systems still presume a vehicular environment that's mute. And that's one thing that will change over the next 10 years. Several initiatives around the world are considering standards for vehicle-to-vehicle (V2V) communications, in which new cars would be fitted with low-cost, short-range wireless transmitters. They would continuously alert surrounding vehicles (as well as elements of the highway infrastructure) to the vehicle's trajectory, the driver's actions, perhaps even the car's ultimate destination. The infrastructure, in turn, would alert cars to accidents, congestion, speed zones, vehicles nearing crossroads, and other conditions ahead.