Throughout history, science fiction writers, futurists, and frustrated motorists have fantasized about self-driving vehicles. However, the road to utopia is paved with more than a few technology potholes. Although many of today’s vehicles offer sophisticated assisted driving features—adaptive cruise control, automatic braking, lane departure warnings and self-parking—assembling all the pieces of the technology to build a completely autonomous vehicle has proved nothing short of daunting.
"We have autonomous functions but we do not yet have an autonomous vehicle," states David Alexander, senior research analyst at Boulder, Colorado-based Navigant Research.
This is about to change. Researchers and automakers are now developing fully automated vehicles that can recognize traffic lights, read road signs, navigate via sensors, and perform all the necessary safety checks required for robotic driving. A person would enter a destination into the vehicle’s navigation system and it would then travel to that destination—either occupied or unoccupied—without human involvement or control.
Such technology may arrive sooner than imagined. Since 2010, Google has operated a self-driving automobile that uses a 64-beam laser system to sense what is happening around the vehicle. The car (actually a test fleet of 10 modified vehicles from Audi, Lexus and Toyota) has traversed San Francisco’s curvy and steep Lombard Street and negotiated the Golden Gate Bridge, as well as roads surrounding Lake Tahoe in California.
Meanwhile, research firms such as Navigant and KPMG are predicting that self-driving vehicles will appear commercially between 2020 and 2022. "The technology and conceptual framework are already in place," Alexander says.
The goal of developing an autonomous vehicle is rooted in safety and a more efficient use of highways and byways. Self-driving cars could eliminate risks from drunk and distracted drivers; purge freeways of speeders, tailgaters, and chronic lane-changers, and make driving safer for seniors and the disabled. "The ever-increasing mass of safety features on cars is there largely to protect us from our own stupidity," states Ian Riches, director of the Global Automobile Practice at U.K.-based Strategy Analytics.
In fact, human error accounts for 70-80 percent of collisions, according to the U.S. Department of Transportation. The World Health Organization reports that 1.24 million road traffic deaths occur each year.
Self-driving cars could also usher in costs savings related to operating vehicles and maintaining infrastructure. "It would be possible to drive much closer to other vehicles while optimizing fuel consumption," explains Mark Campbell, director and chair of the School of Mechanical Engineering at Cornell University and leader of a research group that develops autonomous systems.
Experts say that autonomous vehicles would likely arrive in two phases. Initially, the technology would provide drivers with automated features they could switch on and off. Later, always-on automated cars—featuring networked vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2X) systems—would drive bigger changes, particularly surrounding the way people think about cars—and lower car ownership costs. A 1995 study conducted at the University of Southern California found that creating so-called road trains—a concept Volvo has demonstrated successfully—could improve fuel efficiency by 30 percent.
"Self-driving cars could create more of a mass transit mindset rather than the current ownership model," Campbell says. For example, society could adopt widespread car sharing; an individual could order a vehicle using a smartphone and have it arrive in minutes. Automated systems could also make it possible to eliminate the need to manually park a vehicle; a person or group could step out of a car in a passenger zone at an airport or shopping mall, and the car would park automatically and later return on command.
For now, researchers and engineers are busy constructing computer models and running simulations to understand everything from different types of turns to how cyclists and pedestrians behave on different roads and under different weather and traffic conditions. "The challenge is combining models revolving around the physics of automobiles with the computational aspects of driving, including how different objects behave in different situations. The algorithms must incorporate a great deal of probabilistic data," Campbell points out.
Much of the research focuses on a so-called "density function" that incorporates the most common data and most likely outcomes. However, over time, autonomous vehicle systems will learn and evolve. Riches says that the primary technical challenge is building computational models and systems that address the dizzying array of real-world variables and issues, and address the relationship between autonomous and non-autonomous vehicles.
There’s also a need for flawless mapping, Riches says. "It’s one thing to produce a vehicle that can be fully autonomous for 98% of situations. That last 2% is literally the killer."
Yet autonomous vehicles are steadily creeping toward commercial development. Google’s self-driving vehicles have a flawless record in driving over a distance of more than 700,000 kilometers (435,000 miles). BMW, General Motors, Toyota, Volkswagen, and Audi have all acknowledged they are developing autonomous cars. GM has reportedly built a prototype vehicle, while BMW has tested a semi-autonomous vehicle on Germany’s autobahn. Nissan has pledged to have "multiple, commercially-viable Autonomous Drive vehicles by 2020."
Meanwhile, Daimler AG has successfully tested an "intelligent drive" system on a Mercedes-Benz research vehicle. The company is now preparing the technology—which uses sensors, cameras, stereoscopic cameras and LIDAR—for actual deployment.
Shifting Toward Production
The first autonomous driving systems will be available on luxury vehicles at a price premium of several thousand U.S. dollars. Navigant Research estimates that sales of fully autonomous cars will swell from 8,000 units per year in 2020 to 95.4 million per year in 2035, when they would make up about 75% of all light-duty vehicle sales. However, automakers are currently working to lower the cost of the technology. For example, Google’s LIDAR system currently costs about U.S. $70,000.
Alexander says that, initially, there’s likely to be some resistance to autonomous vehicles. A J.D. Power survey found that 37 percent of motorists view the technology favorably, and 20 percent would spend U.S. $3,000 for an autonomous system in their next car. A stumbling point, the firm noted, is trust and confidence in the technology. In addition, it found that motorists want to have the option to drive manually during "boring" driving, "pleasure" driving or for special maneuvers.
Legal issues also present obstacles. For example, in the U.S., only Nevada, Florida, and California currently allow self-driving cars. Alexander says autonomous vehicles will also require entirely new regulations, and significant changes to licensing rules. What’s more, society must establish a legal framework, including who is liable in the event of a collision.
Nevertheless, autonomous cars aren’t far down the road. The technology could automate trucking, eliminate taxis as well as the need for a second car, drastically reduce road barriers and signage and, in a V2X environment, create dynamic lanes on roads and freeways. As Riches explains: "It’s possible to have six of the eight lanes going into town in the morning and six of the eight lanes leaving in the evening. Let the software sort it all out. The driver won’t be confused."
Samuel Greengard is an author and journalist based in West Linn, OR.