Source: Wired, Steven Levy
Photo: Stephen Lam
I went for a drive in San Francisco’s Mission District last month. It was late morning, and there wasn’t much traffic. As I wended my way through the side streets, I avoided a double-parked armored car and steered around construction sites. Though it might have seemed like an aimless outing, my brief sortie was anything but. Every centimeter I drove, every object I encountered, and even the double line I crossed to avoid the Brinks truck was being recorded by a device affixed across the top edge of the windshield, just above the rearview mirror.
Soon, thousands of people might be installing those gadgets in their cars, hoping to make some extra bucks—and, in the process, contributing to the next great crowdsourced project: a ridiculously detailed and constantly updated map of the world’s roads, readable only by the vast swarm of self-driving cars that will populate our byways.
The device is made by a San Francisco-based startup called Mapper, which comes out of stealth today after a year of development. The company’s maps don’t resemble the classic gas station fold-outs, or even the ones made by Google or Apple that have supplanted them. They are meant for machines, not humans, and when you see them rendered, they are made up of glowing pixels where objects, lane markers, and traffic signals are delineated by rough shapes and tell-tale colors. These are the maps of the future, and allegedly the bedrock of a multi-billion-dollar market. Self-driving cars can’t operate without such maps.
“As humans, if we are blindfolded and dropped in a new place, we’ll find our bearings—we have millions of years of common sense to help guide our awareness,” says Nikhil Naikal, Mapper’s CEO. “A machine, on the other hand, needs a large amount of up-to-date 3D map data to have foresight of what to expect around the corner. And that’s exactly the kind of maps that we deliver.”
Currently, companies that are testing autonomous vehicles—such as Waymo (spun out of Google’s research division), Uber, General Motors, and others—have to make their own maps. It’s a painstaking process that requires people to drive vans equipped with sophisticated lidar (a combination of lasers and radar) equipment over designated roads in multiple passes to log every curb height, fire hydrant, and lane marking. As a result, those vehicles are virtually fenced in by the pre-mapped region; a Waymo car’s self-driving mode won’t even kick in unless it senses that it’s in a mapped zone.
Mapper’s solution is to create an army of part-time workers to gather data that will accrue to a huge “base map” for autonomous cars, and to update the map to keep it current. Think of the work as an alternative to driving for Uber and Lyft, without having to deal with customer ratings or backseat outbursts from Travis Kalanick.
The key to Mapper’s scheme is that it can create high-definition 3D maps without using lidar. That expensive and sometimes finicky combination of lasers and radar has become the standard not only for piloting autonomous cars, but also for producing the cartography that grounds them. Mapper’s founders are PhD engineers who have participated in DARPA challenges and created an indoor mapping startup, Flyby Media (bought by Apple), and the company believes it can match lidar-quality results by relying on sophisticated modeling and data-compression techniques that allow it to use over-the-counter parts to snare data.
The company has created a femur-sized plastic device called the S1, which has multiple cameras and sensors that goes over one’s dashboard and a single cable connecting it to the cigarette lighter for power. It wirelessly syncs with a driver’s iPhone. “It cost $350 to make, it’s composed of commodity parts, and it is designed so it can be easily installed in any car,” says Naikal. After installation, the Mapper app directs the driver to a predetermined route and tells him or her how quickly to go. More accurately, how slowly to go—Naikal says the optimal velocity for a mapping session is between 10 and 30 miles an hour. (Mapper might consider issuing drivers bumper stickers identifying why those cars are creeping.)
Mapper doesn’t expect its drivers to work full time; Naikal says that more than four hours following the exacting directions from the app results in “cognitive overload.” Ideally, his drivers will use the app for an hour or two at a time. In addition, when not taking directions from the app, drivers can leave the system on, and Mapper will collect the data from wherever they wander. (This passive mode also pays, but much less.) “This can be anyone who wants to have a side hustle, who wants to make a little bit of money on the side,” says Naikal. “It could be construction workers; it could be people that are just out on the street a lot, who want to have an ability to do something cool and get paid to do it.”
Because a lot of the value that Mapper provides lies in updating the map—city streets might need refreshing at least once a week to reflect construction, new traffic patterns, and even grass growing higher—the work will be consistent. Mapper’s chief product officer, Jonathan Glanz, estimates that when the company scales up, it will be able to maintain its base map with about 10 thousand mappers—far fewer than Uber’s hundreds of thousands. “We won’t need a humongous fleet because we can select the tasks for mapping,” says Alonso Patron, Mapper’s chief technology officer.
Once it creates the base map and keeps updating it—“like a living organism,” says Naikal—Mapper can license the product to customers such as automakers, transportation services like Uber, and even technology companies like Apple and Waymo. What all of those customers have in common is that they are currently paying to do their own mapping, and would have to spend a lot more money for complete coverage. And, Naikal says, they are reluctant to share data with each other—why would Toyota trust Ford’s data? If you had one map—and evolved it using contracted drivers and other data submitted by the customers themselves—it would be like the Intel Inside of 3D maps, the standard digital atlas for the autonomous vehicles that will be our future chauffeurs. If companies want to customize that map for their needs, Mapper will accommodate them. “We want to have one base map,” Naikal says. “We focus on building that core base layer ourselves, owning the data, making it available for all.” When customers have specific needs they can send their own data to Mapper, which can layer it on top of its base map.
After testing the system in Old Alexandria, Virginia, earlier in the year, Mapper is now busily doing a digital Mason-Dixon on San Francisco, hoping to finish a base map of the city by Thanksgiving. The mapping will move to other cities over the next 14 months and then expand to urban centers overseas. After that, the company will turn to rural areas, eventually hoping to capture everything, so that autonomous cars will have a data cushion wherever they roam.
Once Mapper develops its base map, it can make money not only from the autonomous car manufacturers but also from other customers who might make use of the world’s most detailed guide to roadways: utilities maintaining infrastructure, insurance companies looking for hazards, and, perhaps most intriguing of all, augmented reality ventures that might want to transform your road trip with a stream of personalized billboards. (As for privacy issues, Mapper says that it’s not going to collect information like license numbers or house addresses, though its customers may well do so on proprietary layers they develop on top of the Mapper base map.)
“We feel that there might be use cases that we aren’t even aware of,” says Naikal.
Mapper isn’t the only company with these ambitions. Like a sudden traffic jam on 101, a swarm of competitors has popped into the 3D mapping world, each with its own approach but all hoping to become Rand McNally for robots. “It’s a pretty crowded space,” says James Wu, CEO of DeepMap, which has raised $32 million in funding from primo VCs such as Andreessen Horowitz and Accel. (Mapper won’t reveal its funding, except to specify that none of it comes from potential customers.) Wu’s company also uses crowdsourcing as one of its data-collecting tactics.
A Y Combinator company from this year’s winter batch, Lvl5, also offers drivers some pocket change for mapping—in this case, simply attaching their phones to the dashboard and using the built-in cameras to capture data. Lvl5 CEO Andrew Kouri, who formerly worked for Tesla, says that if he sends enough drivers down a road, after 6 to 12 passes, Lvl5 will have good enough info to use its algorithms to add to the map. It pays drivers between one and five cents a mile, but doesn’t send them on pre-selected routes. “We use a lot of Uber drivers,” says Kouri.
Another company, Civil Maps, uses a relatively costly roof-mounted collection device so that developers in the field can contribute to its crowdsourced base map. (Ford is a funder.) Carmera, whose mission is to “democratize autonomous vehicle data,” partners with companies that operate fleets of cars or trucks. And one of the most formidable players is HERE, the spinoff from the broken pieces of Nokia. It has created a platform called Sensoris that accepts data from multiple partners.
Each of these companies hopes to land in a winner’s circle that might be big enough for only one. Mapper believes it has the unique combination of scale and quality to win. “Obviously it’s in everyone’s best interest to have one player who does this,” says Lvl5’s Kouri. Right now, it’s hard to gauge who’s got the inside track. All are scrambling to snare major customers, particularly automakers. I spoke to an executive at one major potential customer who is considering a pilot project with Mapper. “It’s hard to use cameras to do the work of lidar, but it comes down to expertise and history,” he says. “I don’t know if anyone else can do what the Mapper guys can do. Early results say they can do it.”
If a base map is adopted by Detroit, why wouldn’t the tech companies themselves sign on? For instance, Waymo says that for now its mapping is tied to development, but once its systems mature, it may well consider outsourcing the work.
But while it’s uncertain which company might emerge as the mapmaker who defines the world for machines, it does appear that we all might wind up contributing to it. Though Mapper’s founders don’t say it directly, those 10,000 part-timers who rack up a few bucks following the company’s exacting instructions will dwindle when Mapper starts making use of the built-in, road-facing cameras that will inevitably be standard equipment in the cars of the near future. Essentially, we will all have the technology to perform the functions of Mapper’s S1 windshield device—and these might be turned on by default, so we’ll do the work for free.
To fill in areas that aren’t reliably covered by camera-equipped cars, Mapper could still use contractors to follow its routes and ignore the impatient motorists who tailgate them. And when the era of autonomous cars does arrive, they can be dispatched to perfectly follow Mapper’s directions. No matter who is collecting the data, the robots will get their maps.
Photo Caption: From left to right: Mapper CPO Jonathan Glanz, CEO Nikhil Naikal, and CTO Alonso Patron. Photo direction by Michelle Le.