If an airplane crashed every other day in the United States, killing all 180 people on board, it wouldn’t take long for most Americans to stop flying, right? But that’s about the average number of daily fatalities in the United States due to car accidents, and we continue to drive fearlessly. There are about 33 000 traffic deaths annually in the United States and 39 000 in European Union countries.
Ninety percent of roadway deaths could be avoided simply by making cars smarter, according to Alberto Broggi and Azim Eskandarian of the IEEE Intelligent Transportation Systems Society (ITSS). Their solution: Get rid of the drivers and let the cars drive themselves.
That is not as farfetched as it might seem. Two driverless vans recently drove themselves from Parma, Italy, to Shanghai, China, a distance of more than 15 000 kilometers. The vans autonomously navigated over highways, dirt roads, and desert terrain.
Driverless vehicles were one of the safety innovations discussed during the ITSS annual conference, held in October in Washington, D.C.
General Motors, Volkswagen, BMW, and other automakers, as well as Google and a number of research labs, are working to make driverless cars a reality. The road trip to China was the work of the artificial vision research group VisLab of Parma, Italy. Broggi, an IEEE senior member and president of the ITSS, is a pioneer of machine vision applied to unmanned vehicles and is VisLab’s founder and director.
In 2010 Broggi’s group embarked on a test run of two driverless vans, referred to as the VisLab Intercontinental Autonomous Challenge, or VIAC. At the time the longest trip ever for driverless vehicles, it took three months. The fully electric vans—manufactured by Piaggio, better known for its Vespa scooters—were each outfitted with seven cameras, four laser scanners, a GPS unit, and an inertial sensor suite. Two cameras hanging above the windshield provided stereovision, used for identifying lane markings and the terrain slope. Three synchronized cameras behind the windshield stitched their images into a 180-degree panoramic frontal view. The laser scanners—three mono-beam and one four-plane laser beam—detected pedestrians, other vehicles, and obstacles, including potholes.
Each vehicle also carried three computers. Two of them processed images and laser data while the third integrated all the information and planned a path—which in turn triggered controls for steering, accelerating, and braking the vehicle. Solar panels atop the vans powered the electronics. Software took the large panoramic image in front of the vehicle and identified the lead van—they traveled in pairs—even when approaching a tight turn or steep hill. It also detected road markings and obstacles.
Because parts of the route were not mapped, autonomous planning by the vehicles was not always possible, according to Broggi, so pairs of vans—one with a driver and the other without one—traveled the route together. The lead vehicle was partly manual and partly autonomous. On straight roads or highways it was left to drive itself, following the lane markings, but when decisions about the route had to be made, a driver intervened and defined it. This van also conducted experimental tests on sensing, decision, and control subsystems, and collected data.
The second van was totally autonomous. It used its cameras and navigation system to follow the first; it visually tracked the lead van, planned a path in real time, and generated controls for steering and accelerating or braking. The vans averaged about 160 kilometers per day at a maximum speed of 70 km/hour running four two-hour stretches each. A pair of vans drove for two hours and then were taken off the road to recharge their batteries, using power outlets along the way or, when none were available, diesel generators carried in support trucks. Loaded on a support truck, these cars were then driven to catch up with the two vans that had replaced them on the road.
This other pair then hit the road for its two-hour shift and then was replaced by the recharged vans, and so on. The convoy included three support trucks, which provided a mechanics shop and storage, and four motor homes for sleeping accommodations.
Nearly every collision attributed to driver error—that 90 percent mentioned above—could be eliminated if intelligent transportation technologies like the ones used in Broggi’s driverless vans, added to others already used in some vehicles, could be applied, according to Eskandarian, director of the Center for Intelligent Systems Research at George Washington University, in Washington, D.C., and a member of the ITSS board of governors.
“Car accidents are purely manmade and so we should be able to prevent them,” he says. “We have by developing active safety systems.”
Such pre-emptive systems help the driver avoid collisions. Those already in use include lane-departure warning systems that sound an alarm, electronic stability-control programs that steady a vehicle when it negotiates a curve, and automated cruise control that uses radar or cameras to slow down a car when it gets too close to the one in front. For the most part, the systems become active only when necessary, Eskandarian notes. “But these technologies could someday take a much more active role and allow the vehicle to drive itself,” he says.
Adds Broggi, “These systems will be much safer than human drivers since these cars do not drink, get distracted, or drive under the influence of drugs; they will always react properly.”
While the technological challenges for more intelligent vehicles are being addressed, the cost of such systems remains a roadblock. They’re expensive and so tend to be offered only in luxury cars. But, Eskandarian notes, their prices are slowly coming down.
Price is not the only barrier to more widespread use, however. Governments have not yet required that such safety features be installed in all models, as they have for airbags and seatbelts. And more work is needed on standards, because each active safety system has different requirements designed by different manufacturers to their own specifications.
Eskandarian points out that the IEEE Standards Association is working on related areas, particularly in technologies concerning vehicle-to-vehicle communication and vehicle-to-infrastructure communication for such things as route selection, notification of traffic jams, and other driving- and travel-related information. Already rolled out are the IEEE Standard for Wireless Access in Vehicular Environments that include Resource Manager, Security Services for Applications and Management Messages, Networking Services, and Multi-Channel Operation (IEEE Std. 1609.1-4) and a wireless standard for vehicular environments (IEEE Std. 802.11p).
Meanwhile, the insurance industry has raised concerns about who would be responsible if the systems fail and cause a collision. Who pays the fine—the car’s owner or the software’s maker?
And designers of autonomous vehicles have their own concerns, Broggi says. They need to be aware of nearly every scenario that can happen on a road. One big challenge is programming for the dynamic environments of city streets.
Eventually, drivers must be willing to accept that electronic systems will take over control of their vehicles, Eskandarian says.
“The public needs to have an open mind about the car of the future,” he says. “Fuel-efficient cars capture a lot of attention, but safety systems should capture just as much. Now that some of these active safety technologies are available in luxury cars, we need to speed up their implementation.”