Taking Driverless Cars to the Next Level

Driving is hundreds of minute tasks, woven together in one dance, and more and more of these tasks are being isolated and conquered by automated systems.

by Owen Poindexter / November 25, 2014
Audi’s autonomous TTS. Merging self-driving cars with ride-sharing services could reduce vehicle traffic by 90 percent, according to a new report. Audi

For all the futuristic visions that self-driving (or “autonomous”) cars inspire, their progress has been incremental to the point that we barely notice how much automation is already around us. Cruise control has been a standard feature in any mainstream model for over a decade. Emergency automatic braking worked its way into car design from high-end models on down. Electronic stability control, which detects when the car is skidding and brakes specific wheels to make corrections, has quietly become ubiquitous. Lane correction, a system that guides the car back to the center of the lane when the driver starts to drift, has started on the path from luxury to mainstream with little fanfare, and self-parking cars have been on the road for years, albeit in limited models. The day when we barely touch the pedals or steering wheel is coming—eventually—but the actual progress toward that day has largely escaped attention.

It’s difficult, when talking about autonomous cars, to avoid slipping into a sort of techie-utopia: Millions of cars off the road, traffic accidents nearly vanquished, shared cars picking people up and dropping them off, finding a parking spot—a huge source of emissions and wasted time (and brain cells)—largely a thing of the past. This vision relies on what the National Highway Traffic Safety Administration calls “level 4” vehicles: fully autonomous cars that don’t require a driver. Once those are widely available and affordable, we can start to restructure the relationship between humans and automobiles, with more shared vehicles and adjustments to infrastructure that take advantage of automation.

“Automation could be the catalyst that brings together personal mobility and public transportation,” said Sven Beiker, of the Center for Automotive Research at Stanford (CARS).

For the more foreseeable future, however, our cars will generally still require an attentive (if not always active) human driver.

Level 4 autos understandably generate the most excitement, but this has obscured the considerable progress made already on levels 2 and 3, where the car provides assistance, such as lane keeping (level 2), or largely handles the driving in a sort of autopilot mode (level 3). Level 1 refers to cars with no automated functions. Virtually every major manufacturer either has or will soon have a car on the market with adaptive cruise control, which enables the car to automatically adjust its speed to stay a safe distance behind the car in front of it, potentially simplifying highway driving considerably. Lane correction is working its way into more models, and certain luxury vehicles, like Tesla’s Model D, can read and obey speed limit signs. Self-parking is steadily becoming more available.

Driving is hundreds of minute tasks, woven together in one dance, and more and more of these tasks are being isolated and conquered by automated systems. In this way, we have been creeping toward level 3 cars, which require a driver to be ready to take over, but outside of inclement weather or unique traffic situations, the car does the work. The next targets include automatic lane changing, to safely zip around slower drivers, and automated driving in traffic jams. City driving and highway driving are, for now, separate targets for automation, and it’s likely that certain autonomous vehicles released in the next few years will be able to do one but not the other.

“Now we are focusing on scenarios where you take the car to the highway and let the car take over, then you take over again when you get closer to work,” said Marcus Rothoff of Volvo.

Google has announced that they will have cars with autopilot functions for sale by 2017, and it’s important that there is a legal infrastructure that sorts out permission and liability by then.

“It may be that we have level 3’s [in two to three years] but policymakers may slow things down as we get close to full automation,” said Prof. Kara Kockelman, of the University of Texas at Austin.

And yet, unrestricted level 4 vehicles and their ability to transform how urban areas operate are still a long way off.

“We understand automobiles as something that people can take anytime and go anywhere they please,” said Beiker. “But the fully automated self-driving cars might not be able to do this in the same way. We’ll get that when we have flying cars or hydrogen cars.”

For those for whom the flying car is a cruel joke about what futurists promise, Beiker is not saying that we will never have fully autonomous cars, just that they will take decades before we trust them on public roads. Once you add restrictions by area (e.g. within an office park), then the timeline shrinks considerably. A car in a closed environment, like an airport, or a limited environment, like a designated downtown area, can reduce the number of variables the manufacturers need to worry about. This allows for greater assurance with accurate mapping and key features such as good lane marking and devices that can exchange information with cars. These physical elements are referred to as the “built infrastructure” and should be a central point of collaboration between automotive manufacturers and municipalities or large private areas.

“If we say, automated vehicles can operate only in this lane, or we have an automated shuttle service in a downtown area—that can work very well,” said Beiker.

Volvo, for instance, is limiting its initial public release to vehicles in and around Gothenburg, Sweden. Level 3 cars are still being confined to limited environments, but those environments now include public roads in Sweden and California, where Google’s cars have logged over 700,000 miles.

“This evolution from level 2 to 3, then 3 to 4 is not hard to imagine,” said Kockelman, “with increasingly higher speeds and more complex locations with more complex weather.”

Still, what’s available will always lag in years and power behind what’s possible. Auto manufacturers may be deemed liable for accidents that occur while the car is in control, not to mention the press hit any manufacturer would take with any serious accident, so the companies working on these cars are at least as cautious as the people driving them.

“The biggest technical challenge is that the car can trust what it sees it around it,” Rothoff said. A human knows automatically that a person riding a bicycle is potentially hazardous, while a bike locked on the side of the road is not, whereas a car has to learn that. As for the human intelligence employed by drivers, cars are not close to matching that, and it’s unclear if they ever will be.

“To really understand, that’s not just a car, that’s a car driven by a potentially reckless person,” said Beiker, “and that car has way more horsepower than the driver has brain power—a good driver senses that. And since we will hold these vehicles to very high standards, we want these cars to drive better than the best humans.”

Better sensors, highly detailed mapping and vehicle-to-vehicle communication will drive—or inhibit—the climb toward level 3 and eventually completely autonomous cars. For now, these autos will have to narrow their deployment to specific boundaries and opportunities. Areas that already are contained physically and legally, like office parks, universities and retirement communities, will be able to make specific accommodations for self-driving cars, which would likely be owned or leased by the presiding organization.

Automation is already here, and its growth may remain subtle for some time. Like the rise of the Internet, autonomous cars may seem to emerge suddenly, and only in retrospect do we see that the movement had been growing for decades.

This story was originally published by FutureStructure