The Google map on your smart phone is a map made for human consumption. It resembles, in simplistic strokes, the world right before your eyes: road straight ahead, side streets to the left and right, intersection approaching. It talks to you in terms you can appreciate: "turn right in 200 feet." It flags your destination with a big red dot, so even the most directionally challenged can't miss it.
"That information is really intended for you," says Ogi Redzic, senior vice president of "connected driving" for the Nokia-owned mapping company HERE. "The computer processes it, but it then passes you the information, and you actually do something with it. You turn the steering wheel, you go left, you go right."
This isn't a terribly novel observation — navigation maps are meant for people! — until you consider what will happen when people don't do the navigating any more. In a future when driving becomes more automated, maps will be read by cars.
That prospect raises a ton of technical challenges for the engineers already preparing now for a world of driverless vehicles. For instance, you can supplement what Google Maps tells you with your own judgment and eyesight. A ballpark direction — "turn right in 200 feet" — is enough information for you to work with. Not so for a machine. Map errors, meanwhile, that might be amusing to a human — say, when Apple routes you through a cornfield — could be devastating to a car.
"When we say 'turn right,' that decision needs to be made much more precisely. It’s likely going to be 201 feet, 199 feet," Redzic says. "The ultimate consumer of the experience is the human, but the ultimate consumer of the data is the car."
HERE is one of the companies already trying to build such maps meant for machines, so that they'll be ready when automakers roll out cars with more automation. Driverless cars will of course have exterior sensors, too, like the hardware atop the Google autonomous car that's gotten a lot of attention. But they'll need these kinds of internal mapping brains, too, not just to navigate from point A to B, but to know precisely where they are and what's coming up that sensors can't see.
"A bad snow day — that’s also going to make sensor-based decision-making much more challenging," Redzic says. "We always view map data as an extended sensor, as allowing sensors to see much further into distance."
These maps will first and foremost have to be a lot more precise than what we're used to seeing on smartphone apps or in-car navigation screens. This picture, for instance, shows the difference between roads as they're mapped on traditional maps that may be precise down to the meter (in green), and roads as HERE is trying to map them down to the centimeter:
When cars are processing this information, each lane must be individually modeled. The car will need to know that you can't turn left from the right lane (or right from the left lane). Such maps would have to include the slope and precise curvature of each little bit of roadway, as well as the elevation (so that, for instance, a car knows where it is and what to do here). These maps would have to convey which roads travel one-way at rush hour, and exactly when, which intersections have stop signs instead of stop lights, which streets have bike lanes and on-street parking. They'll need to know the exact width of each lane, and every change in lane markings.
Oh, and they'll have to be constantly updated, because an outdated map detail is a big safety problem.
That means whatever mapping data HERE or companies like it produce will rely on constant feedback from cars on the road. Maybe you'll buy a driverless car one day that uses sensors to find its way, but also to measure changes in the landscape to update these maps. This is, in fact, how HERE builds these maps today: It has vehicles — they look about as silly as the Google Street View car — that map roads with LiDAR sensors as they travel.
The result, for now, looks something like this, a high-definition, 3D map that gives a whole lot more information than the Google Map above:
This picture raises another question: What will happen when your driverless car drives to a place that hasn't been mapped yet? HERE won't have the entire world mapped in this way by the time the first fully autonomous car comes out.
"I do not envision a scenario where the first car will be able to drive anywhere in the world," Redzic says.
You'll ride to, say, the Mexican border, and then you'll have to stop there, beyond which, the world — to a car — will be literally off the map.