Driverless Cars Need Just One Thing: Futuristic Roads

Driverless cars have moved with remarkable speed from DARPA-funded fantasy to picking up passengers on the streets of Pittsburgh. The excitement is justified, in part, because there’s much to gain. A single, shared autonomous vehicle could replace roughly 11 privately-owned vehicles, according to a recent University of Texas study. By reducing the number of cars on the road, self-driving vehicles could cut traffic, emissions, and urban sprawl, while improving safety and saving money for the millions of households that would no longer have to own a vehicle.

Yet we’re still a long way from adopting a futuristic fleet of driverless vehicles, and the main obstacle is navigation. Manufacturers teach their cars to move by employing fleets of drivers who travel the streets in ordinary cars, scanning for changes in previously mapped roads. These changes are pushed, once a month or so, to driverless vehicles, so that the cars can interpret the roads correctly as they drive.

This works for a lot of what vehicles encounter. Some things—like streets, curbs, and intersections—don’t change for years, while other, temporary things—like traffic delays and road closures—can be crowdsourced by mapping devices, such as Waze. But driverless cars need to “see” some information that’s blocks or miles away in real time—such as a dog running into the street, a speeding ambulance, or a swerving driver.

There’s a way to get this information to vehicles quickly and accurately. Just as lawmakers and city planners started laying asphalt, installing streetlights, posting speed limits, and zoning property to accommodate Henry Ford’s cars, we need to design roads and infrastructure for the self-driving generation of vehicles.

Building roadside sensors along our streets and highways so that cars can navigate them sounds, initially, like an expensive pipe dream. But a series of advances are coinciding to make such a system possible. Fifth Generation wireless, or “5G,” is a wireless broadband standard that’s currently under development and will improve on our current networks in ways that will be useful for driverless vehicles. A 5G system will include denser wireless networks (with facilities every few hundred meters in cities) speeding up connections. 5G will also have broadcast capabilities, allowing networks to distribute large amounts of mapping data without time lags.

Sensor technology is also getting better and cheaper. To sense and react to the world, driverless cars use a series of technologies, including radar, cameras, and lidar. Lidar systems emit invisible laser light that reflects off of objects in view; then, the lidar system times the speed of the reflection to measure distance, creating a precise, three-dimensional image of the world. The best systems don’t come cheaply. The distinctive spinning lidar unit that you see on top of Google’s automated Lexus SUVs cost about $75,000 per vehicle.

But the price of high-quality systems is falling rapidly. In April, a Silicon Valley-based company released a $500 lidar sensor that’s under two pounds. Ford is using these smaller devices on its driverless cars, which will hit the roads later this year. Low-power lidar units are improving and cost under $100. Some in development are smaller than a coin and may one day provide low-cost mapping to supplement the information captured by car-mounted sensors.

Mounting these sensors to cars comes with drawbacks, especially for vehicles that are only partially autonomous. Car-mounted lidar sensors have a range of 200 meters, and perform best under 100 meters. That means that if a car is traveling between 30 and 60 mph, a passenger will have around 5 to 15 seconds of warning of an unusual situation up ahead.

These takeover requests can be jarring, irritating, and dangerous. Researchers at Virginia Tech found that passengers took an average of 17 seconds to begin driving in such situations. Passengers on road trips, who might be sleeping, eating, or watching a movie would need even more warning. Roadside sensors, however, would allow vehicles to “see” activity far ahead on their routes. With roadside sensors, passengers might have minutes, instead of seconds, of warning. Smart roads could also provide more descriptive information to vehicles, letting them know, for example, that the person-shaped object near a sensor is a Halloween scarecrow, not a child.

There are other benefits. Car-mounted sensors are often confused by road materials (a shift from dirt to gravel or asphalt), reflective buildings, bridges, or even the weather. Roadside sensors not only mitigate these problems, but also reduce the computing load on car-mounted systems, because the vehicles have to make fewer snap decisions.

The falling price of lidar means that blanketing a city or highway is relatively inexpensive. Take Washington, D.C., which has 1,500 miles of public roads. For roughly $24 million, a firm could purchase enough lidar devices to have readings of every public road, with a device every 160 feet. Lining the 430-mile stretch of I-95 between Boston and Washington, D.C. would cost roughly $4 to $15 million, depending on spacing. It sounds expensive, until you compare these figures to the $200 million Washington, D.C. spent on its 2.2-mile long streetcar system.

A system of roadside sensors would require infrastructure. Cities and counties have to bury protective ducts for wires, construct poles, and permit wireless sites on public property. In large cities, which typically have decades-worth of underground structures such as subway grates and sewer lines competing for space, it gets tricky. This urban infrastructure is often jointly owned and regulated; gaining permission to dig up streets and attach thousands of roadway sensors would be time consuming and expensive.

But over the next decade this infrastructure is going up anyway, as cities prepare to offer Americans high-speed internet access. Overlaying a roadside sensor system would be relatively easy. And in rural and suburban areas, where fewer approvals are necessary, such a system would be have a greater impact. Semi-automated vehicles might provide an alternative to public transportation, in areas replete of a reliable system. Crowd-sourced mapping isn’t as effective in rural areas, as there’s not enough traffic for real-time updates, making roadside sensors the only reliable option for self-driving cars.

This is not to say that self-driving cars are the only thing to be benefitted by deploying sensors to our roads. Cities could use sensor data for conducting traffic studies, pushing out real-time public bus alerts, increasing parking space occupancy, metering commercial loading times to prevent congestion, and enhancing pedestrian safety. There are also commercial applications for sensor data: How many cars drive by a billboard? How many people walk by a storefront per day? How many of those people have dogs? These are all questions we could easily answer with roadside sensors.

One of the main reasons firms shied away from earlier plans for sensor systems was the extensive government intervention involved. Regulators and legislators should not prescribe technologies. As the evolution of the internet, operating systems, and cellular phone standards show, there are tremendous social benefits when firms are freed to iterate and compete. But the installation of roadside sensors requires forward-thinking state and municipal governments, who have the legal authority over most rights-of-way, protective conduit deployment, and wireless siting. They can help usher in this transportation revolution.