Click to enlarge picture
GM's EN-V robotic concept vehicles rely on communication with other vehicles to drive themselves safely.
Google's landmark deployment of autonomous cars — vehicles that can drive themselves — onto public roads and into live human traffic last summer and early fall was announced casually, and after the fact.
Astonishingly, the search-engine giant was able to set loose seven Toyota Prius hybrids, all adorned with a dizzying array of odd-looking sensors, onto Highway 1 between San Francisco and Los Angeles for several months without raising suspicion. Each vehicle was piloted by artificial-intelligence software designed to interpret the data collected by the sensors and use it to mimic the decisions made by a human driver. The goal: to fundamentally change the way we use cars.
How so, you ask? Google believes that the use of autonomous vehicles could nearly halve the number of automobile-related deaths — which it estimates at 1.2 million worldwide per year — because computers are theoretically more precise drivers than humans. In addition, the instant reaction time and 360-degree awareness of computer-controlled vehicles would allow them to ride closer together on the highway than vehicles driven by humans, thus reducing traffic congestion. And finally, they can be more fastidious with the accelerator, reducing fuel consumption and carbon emissions considerably.
Essentially, riding in an autonomous car could shave time off your daily commute, reduce your carbon footprint, save you money and save lives in the long run. And you don't even have to lift a finger. Instead of driving, you're a passenger — working, watching television, conversing with friends. Sounds idyllic.
Does this mean that self-directed robot cars, the kind that science-fiction writers have been dreaming about for decades, will hit the streets within a couple of years? No.
While the Google project may be one of the most high-profile demonstrations of autonomous-vehicle research, and one of the most successful to date, the path to a production robotic car still remains uncertain and would require clearing a staggering number of technical and legal hurdles. But plenty of people, in both the academic world and in the research-and-development divisions of carmakers such as GM and Volvo, are working out the kinks.
Bing Images: Driverless Cars
Safety in Numbers: Networked Autonomy
The basic recipe for the robotic car was developed through trial and error nearly a decade ago: Radar detects objects at long range; laser range finders (also called laser radar, or lidar) sift through those and other objects as they get closer; vision cameras look for specific types of objects; and a bank of computers collates all of these data and tells the fully electronic, drive-by-wire vehicle where to go.
But so far no one has ever been able to create a vehicle that can handle the vagaries of everyday life on the streets.
Even the most headline-grabbing accomplishments of robotic cars, such as Stanford University's driverless Audi TTS, which recently climbed Pikes Peak, were highly controlled, precisely plotted experiments, with little bearing on the chaos of real-world driving. How do robots respond to a blizzard, to drivers who won't yield or to something as routine as a detour? Being able to recognize, read and process a blinking detour sign, for example, is a feat of artificial intelligence that's still pure science fiction.
One interim solution is to limit the burden of autonomy by programming robots to follow a human leader. The goal of the European-based Sartre project (Safe Road Trains for the Environment) is to create road trains, harnessing existing driver-assistance features to establish a convoy of robots behind a human-controlled truck. Unlike most autonomous vehicles, which bristle with mounted sensors, these are production-model Volvos that use factory-standard sensors, vision processing and electric-motor-assisted steering and braking to stay in position.
"We're trying to stretch the existing technology as far as it can go," says Erik Coelingh, technology leader for active safety at Volvo. Avoiding some of the more expensive standbys of autonomous vehicles, such as the ubiquitous laser range finders bolted onto most robot-car roofs, means working with lower-quality data. So the only aftermarket add-on that Sartre vehicles use is a wireless communication system that keeps the cars in contact with each other and coordinates vehicles entering or leaving the convoy. Sartre has been tested at low speeds, with one lead vehicle and one follower, but Coelingh plans to test a full convoy this year, eventually reaching speeds of 55 mph.
General Motors is also developing networked autonomy with its collection of concept vehicles, called the Electric Networked Vehicle, or EN-V. In theory, the podlike EN-Vs would access a citywide or areawide network and coordinate their movements as kind of cooperative robotic fleet.
GM has conducted autonomous-ride events, most recently at the Consumer Electronics Show in Las Vegas in January. But as EN-V project director Chris Borroni-Bird points out, there's a difference between weaving around an enclosed space and navigating city streets: "The network has to be completely stable. No dropouts, no GPS loss of signal and no wireless loss of signal. It's one thing when your cell phone loses its signal. It's another if the vehicle is driving autonomously and suddenly doesn't know where it is."
Read: GM Unveils Future City Car in Shanghai