Thanks to advanced driver assistance systems (ADAS) ranging from adaptive cruise control to lane keeping assist and automated steering, today’s automobile is very nearly autonomous. The driver remains in control, of course, but built-in sensors, cameras and radars make it possible for the vehicle to take over much of the actual driving task.
With new technologies, the fully autonomous automobile will one day be widely available. It’s just a matter of time, experts say, for the concept to move from a research and development exercise to full-scale production.
The Benefits Are Apparent
Human error is overwhelmingly to blame for the vast majority of automobile accidents today according to statistics cited by autonomous vehicle experts, and the economic cost of these accidents has been rising.
For example, says Egil Juliussen, principal analyst for infotainment and ADAS at IHS Automotive in Chicago, more than 90 percent of traffic accidents are caused by human error, and in 2010, traffic accidents in the U.S. cost more than $300 billion, up from more than $230 billion in 2000. Moreover, Juliussen notes, this cost increase has occurred even as accident rates in the U.S. have been falling.
By contrast, self-driving cars are expected to save lives rather than put them at risk.
In October 2013, the Eno Center for Transportation, a non-partisan think tank based in Washington, D.C., released a research paper called Preparing a Nation for Autonomous Vehicles: Opportunities, Barriers and Policy Recommendations, which estimates 1,100 lives would be saved and there would be 211,000 fewer crashes per year if 10 percent of vehicles on U.S. roads (12.7 million) were autonomous. With 50 percent, or 63.7 million autonomous vehicles, the estimates jump to 9,600 lives saved and 1.88 million fewer crashes. With 90 percent, or 114.7 million autonomous vehicles, the Eno Center projects 21,700 lives saved and 4.22 million fewer crashes.
Google, which has been testing driverless vehicles since 2009, now has about two dozen self-driving Lexus RX450h sport utility vehicles (SUVs) traversing roads nationwide. To date, they’ve collectively logged more than 500,000 miles, and none has caused an accident while in self-driving mode, the company says.
Additionally, self-driving cars are inherently more efficient than human-driven cars, and thus could yield time savings and environmental benefits, the Eno Center notes. Because a self-driving car is better at gauging and utilizing roadway space, it could help to reduce traffic congestion. So could the autonomous vehicles’ safety benefits alone help reduce congestion, the Eno Center says, citing Federal Highway Administration research that attributes 25 percent of traffic congestion to crashes and other “traffic incidents.”
Reduced traffic congestion would also lower fuel consumption, ultimately lowering the vehicle’s greenhouse gas emissions.
Self-driving cars also could offer new freedom to the elderly and disabled who are unable to drive a vehicle themselves.
So, with the benefits of self-driving cars clearly apparent, research and development is expanding. Automakers including Audi, General Motors, Hyundai, Mercedes-Benz, Nissan, Tesla and Volvo are working on self-driving cars. And auto industry suppliers such as Continental Automotive Systems are working to develop necessary technologies, too.
Advances Will Be Gradual
“For the foreseeable future, the human has to be in the loop,” says Raj Rajkumar, professor of electrical and computer engineering at Carnegie Mellon University in Pittsburgh. “Autonomous technology is a lot more mature than what people think, and it’s happening much faster than anyone predicted,” he says.
CMU researchers including Rajkumar have been working on self-driving cars since the mid-2000s. In 2007, a team from CMU won the DARPA Urban Challenge, an autonomous vehicle race organized by the U.S. Defense Advanced Research Projects Agency that involved driving 60 miles over a six-hour period on paved and unpaved roads at a decommissioned Air Force base. Of 125 initial teams competing, six completed the race, and their vehicles had to obey mock traffic rules in realistic scenarios—including red, green and yellow traffic lights at four-way intersections—divert themselves around blocked routes, and interact with fixed and moving obstacles such as other vehicles.
“It was a watershed moment,” Rajkumar says. “It was the first time it was demonstrated that vehicles could drive themselves, without anybody in them, following all the rules that you and I follow when driving—the first time in history autonomous driving in realistic conditions was no longer science fiction.”
Nevertheless, the auto CMU modified for the DARPA Urban Challenge was a 2007 Chevy Tahoe SUV that “looked like a science project,” Rajkumar says, with sensors conspicuously adorning its exterior and lots of additional electronic equipment inside the cabin.
Since 2011, the university has been working on a completely new platform, a model year 2011 Cadillac SRX crossover utility vehicle (CUV) that looks completely normal inside and out, having been modified with automotive-grade cameras, radars and light detection and ranging (LIDAR) systems that are like those now available factory-installed in many newer car models.
This was a technical milestone, too, Rajkumar says, because it proved that a normal-looking autonomous vehicle could operate in a high traffic density suburban environment with traffic lights and very complex intersections, as well as in a relatively simple highway environment.
Fully autonomous vehicles will be available in the 2020s, “when you could be reading a newspaper” while the car does all the driving, Rajkumar says.
“It’s a ‘when’ question, not an ‘if ’ question,” and the advance to full autonomy will come in multiple stages, says IHS’s Juliussen. In the first stage which extends through at least the next five years, he says, “the human can take over and drive.” But much later, around the year 2030, he speculates, there will come a stage when the human cannot take control, because the human-machine interface will have been removed, leaving only the machine to handle the driving task. Then everyone in the vehicle will be a passenger; there will be no driver.
By that time, he predicts, the advantages of self-driving cars will be self-evident: lower vehicle cost, improved safety, and greater convenience.
Juliussen notes that human-machine interfaces are expensive and that car prices will drop when you no longer need to add these complex systems to cars. “The cost of all the electronics and the software to perform the self-driving initially is going to be pretty high,” Juliussen says. But he adds that “as production increases and technology advances, that will drop fairly nicely.”
According to the Eno Center paper, though, it could take at least a decade for the cost of an autonomous vehicle’s added electronics to fall to $10,000. By comparison, just the LIDAR system on the roof of one of Google’s autonomous vehicles costs $70,000, and the additional cost of other sensors, software and engineering now can bring the total cost of an autonomous vehicle to more than $100,000, the Eno Center paper notes.
“It’s going to be a phased approach,” says Andy Gryc, automotive product marketing manager at QNX Software Systems Ltd., in Ottawa, Ontario, which has supplied technology to power autonomous systems since the 1990s—including technology for vehicles competing in the DARPA Grand Challenge, the agency’s first autonomous vehicle contest, in 2004. “How people define those phases is variable,” he adds, “but generically it’s from the vehicles providing some sort of assistance like adaptive cruise control, so the car can drive itself under limited circumstances (such as automated steering on highways), to autonomous driving under normal circumstances with the driver required to be on call, and then fully autonomous.”
Gryc says the lines between phase changes will be blurry. Two different automated driving systems could appear to be part of one phase or another. “I see it being more of a gradual introduction of technologies. Certainly, by 2020, I think we’re going to see close to fully autonomous vehicles,” he says.
Automakers Pursue Objectives
Currently, automakers are striving to make the leap from level two to level three in production vehicles—a leap that is still five to 10 years away from actualization, says Erik Coelingh, senior technical leader for safety and driver support systems at Volvo Car Corp. in Göteborg, Sweden.
When Volvo introduces its all-new XC90 SUV this year it will be a level 2 vehicle equipped with lane-centering autonomous steering technology that uses a camera to see lane markings, road edges and barriers, position the car in the lane and anticipate road curvature ahead—similar to the lane-centering system already offered in 2014 iterations of Mercedes-Benz’s S-Class and E-Class sedans. It will require the driver to keep his hands on the wheel at all times, even when the vehicle is steering itself.
Level 3 technologies, which Volvo has already begun to develop, will expand on this, freeing the driver to do other things and be productive while the car controls itself. But even then, “there will be limitations” on certain stretches of road and in certain weather conditions, Coelingh says.
Yet, the automaker “cannot count on the driver taking care of all those exceptional situations,” he says. So Volvo researchers also are working to develop better cameras and sensors that can detect everything on the road, including debris; better map data to more accurately position the car on the road and anticipate terrain ahead; and redundant systems that can maintain control of the car when a primary system fails to work and the driver doesn’t respond quickly enough.
“If something breaks down, the car needs to be able to take itself into a safe state of some form and there are different versions of that,” Coelingh says. “But in all versions, it is true that the faster you drive the more difficult it is to get into a safe state, and the consequences of an error are much more severe at high speeds than lower speeds. So I think speeds for self-driving cars in the beginning will be very much limited,” he adds.
To help concentrate its research on relatively simpler, speed limited scenarios first, Coelingh says, Volvo is first focusing on developing level 3 systems for self-parking and navigating highways.
Technologically, the transition to level 3 from level 2 will be “immensely difficult,” Coelingh says. “We don’t have all the answers how to do it, and probably we don’t have all the questions, either.”
Audi is similarly concentrating on parking, highway driving and system redundancy for its first level 3 self-driving cars, says Thomas Mueller, head of development of braking, steering and driver assistance systems at Audi AG in Ingolstadt, Germany.
“Highways are the first step,” Mueller says. They present a “pretty reasonable and not complex situation that we can handle,” and one that helps customers get used to autonomous driving technologies.
Self-driving in cities is more complex and will require more research and development, he says, offering the example of an intersection with three traffic signals for three lanes: one for a left turn, one for going straight, and another for a right turn. While a person can intuitively grasp the situation through contextual awareness, the car could get confused by it, Mueller explains.
By the end of 2014 or early 2015, Mueller says, Audi will be introducing its next electronics platform that will “prepare the car for this highly automated technology.”
With the current state of technology, a fully autonomous vehicle would not be able to deal with a pedestrian who is waiting to cross a street in its path and has signaled the driver to go first, says Dr. Eberhard Zeeb, senior manager of driving automation for Daimler AG (Mercedes-Benz’s parent company) in Boeblingen, Germany. The vehicle would stop and continue to wait until the pedestrian made a move, he says. Training a car to be resilient in all situations, to react like a human, is not possible with current sensors, he adds.
Acquiring high-quality maps to improve “localization,” so the car always is aware of its location, is another challenge, Zeeb says.
Nevertheless, Zeeb says, Mercedes-Benz has proven with its S500 Intelligent Drive research vehicle that the sensors used in today’s production vehicles are sufficient to have an autonomous vehicle drive itself on both interurban and urban roads. In August, the car self-navigated the same 100-kilometer route from Mannheim to Pforzheim driven by motoring pioneer Bertha Benz exactly 125 years earlier. The vehicle used “near-production-standard technology” to accomplish the feat,
Mercedes-Benz said in a press release afterward.
The ultimate reason for autonomous vehicles is to eliminate accidents, says Maarten Sierhuis, director of the Nissan Research Center in Sunnyvale, Calif.
“We’re changing the way technology and people interact,” says Sierhuis, who previously was a senior scientist at the U.S. National Aeronautics and Space Administration (NASA) working on human-robot interaction technology in intelligent space suits. For an astronaut outside a spacecraft, the space suit is his home, his work environment and his life support system. An autonomous car is very much like that space suit, “like a robot I am sitting inside of,” he says. “I am inside the machine—and the machine has to keep me safe, and the machine has to do what I would like it to do, and to do it in a way that I would want it.”
Keeping Drivers Engaged
The advent of fully autonomous, self-driving cars can create wondrous benefits for society—for example, offering mobility to populations of people who may not otherwise have it. But there are caveats, academic researchers note.
It’s true that any benefits from self-driving cars could also be tempered by the limitations of the people riding inside these machines. Work toward perfecting autonomous vehicles must be accompanied by efforts to better accommodate the people who will use them.
“Right now we have all or nothing. You have a driver’s license or you don’t,” says Mitchell Moss, director of the Rudin Center for Transportation and professor of urban policy and planning at New York University. But the driverless car eliminates this dichotomy, Moss says, giving new mobility to both senior citizens who lost their ability to drive, and to children who aren’t yet permitted to drive.
“Long term, we now know that driving need not be done by a driver,” he adds. Parents will no longer have to be chauffeurs for their offspring, and anyone can be more productive during drive time, taking care of business or being otherwise occupied as the car takes over the driving task.
Yet long term is definitely far in the future. “Technological diffusion of this sort takes many years,” Moss emphasizes.
“The reality is that we as humans will remain, at least for the foreseeable future beyond 10 years, the responsible party at the control of that vehicle,” says Bryan Reimer, research scientist at the Massachusetts Institute of Technology’s AgeLab and associate director of the New England University Transportation Center, both in Cambridge, Mass. “Unfortunately, what has been shown time and time again, from auto to aviation to nuclear power and beyond, is that we as humans do a pretty poor job at supervising highly automated systems. We fall asleep, we get bored, we try to do other things and have difficulty maintaining attention.”
So, on the road to fully driverless vehicles research must address the issue of reviving a driver’s attention when the vehicle encounters situations it can’t handle on its own. Reimer says, “Driving is a monotonous activity to start with,” and the self-driving car is “making [it] even more monotonous.” The more reliable the autonomous car is, he adds, the harder it will be for a person to be aware and alert when a problem occurs. “The assumption today is that as we automate more, the driver’s skill will maintain integrity.” But instead, the reverse is true. “As the number of vehicle miles driven diverges from the vehicle miles traveled, skill erosion will occur,” he says. “We’re the threat to the automation.”
Regulators Set Targets
Last May, the U.S. Department of Transportation’s National Highway Traffic Safety Administration (NHTSA) released a policy statement defining five levels of vehicle automation, including four concerning self-driving vehicles.
is a car with no self-driving capabilities; the driver is fully in control of steering, acceleration and braking at all times.
Level 1 provides “function-specific automation” in which the vehicle automatically assists the driver by taking over one or more specific control function. NHTSA gives the example of electronic stability control, which applies braking to keep the vehicle planted.
provides “combined function automation” in which “at least two primary control functions” are taken over by the car and used in unison to “relieve the driver of control.” NHTSA gives the example of adaptive cruise control in combination with lane centering. The former controls both the throttle and the brakes to maintain a safe distance behind a lead vehicle as well as a set speed, when possible. The latter automatically steers the vehicle to keep it in the center of the lane, even around curves when possible.
provides “limited self-driving automation” in which the vehicle can be fully autonomous “under certain traffic or environmental conditions.” It’s the vehicle’s responsibility to continuously monitor for changes in those conditions that would require the driver to take back control and to give the driver a “sufficiently comfortable transition time.” NHTSA points to the capabilities in Google’s automated vehicles as an example.
Level 4 provides “full self-driving automation” in which the vehicle is always in full control of itself and monitoring roadway conditions continuously from the start of a trip to the end. “Such a design anticipates that the driver will provide destination or navigation input, but is not expected to be available for control at any time during the trip. This includes both occupied and unoccupied vehicles,” NHTSA says.
NHTSA is conducting research into self-driving cars and related safety issues with the goal of establishing related standards for when these vehicles become commercially available. It expects to complete phase one of its research in the next four years. The policy statement was intended to set guidelines for states that are looking to pass laws relating to the testing, licensing and regulation of selfdriving cars.
So far, among the states, only Nevada, California and Florida have enacted laws permitting limited operation of experimental self-driving cars on public roadways. All three passed their laws before the NHTSA policy statement was released. “As additional states consider similar legislation, our recommendations provide lawmakers with the tools they need to encourage the safe development and implementation of automated vehicle technology,” said NHTSA Administrator David Strickland in announcing the policy statement.