We’re developing the cutting edge of self-driving technology with advanced localization modeling. It’s part of our growing partnership with NVIDIA.

An autonomous car needs to know where it is at all times.

That sounds obvious, but to get more granular, your car needs more than a simple GPS coordinate on a map. If you’ve ever watched the blue dot on your smartphone app float around in space trying to get a fix on exactly where you are, then you’ve seen exactly why a self-driving car must be a thousand times more precise than standard GPS.

How precisely does a vehicle need to understand it's location? Clearly it's not enough for a car to know it's on a specific road, nor is knowing that the car is on a road headed in a specific direction, updated every few seconds.

To operate safely and comfortably, an autonomous vehicle can never afford to not know where it is. A camera alone might be disrupted by snow or fog. Radar sensors may be blocked by traffic. GNSS may have trouble seeing in dense urban canyons.

To execute a precise turn, to avoid another car, to stop perfectly at the cross-walk, an autonomous vehicle needs its location to be centimeter-accurate. This is achieved reliably by building a car with multiple different sensor types working in concert.

To understand this more, listen to Sanjay Sood, HERE Vice President of Highly Automated Driving.

Today's highly-autonomous cars (and tomorrow's fully-autonomous cars) need to flawlessly accomplish a difficult task. It’s a job that requires an immense amount of computing power. It’s a complex exercise that requires both a rich amount of supporting data, as well as super-powerful processing power that can render decisions in near-real time. Not surprisingly, it’s something that the human brain can do almost instantly.

When you look around your car from the driver’s point of view, you ingest and process an enormous amount of data.

You know which lane you’re in. With a single glance, you sense how far you are from the curb. You know how far the next traffic light is, and when you approach it, you know exactly where to stop the vehicle in relation to the lanes and roads around you.

Enter, the localization model

That’s a lot of information. When an autonomous car performs these same feats, it’s called localization. Proper localization, the kind we’re seeing coming to the road now, covers the range from the macro level of GPS location, down to the micro level of understanding where your car is in relation to the curb, a street sign, and all the elements of the road around you.

To get the localization job done, two things are necessary. The first is a precise, reliable information resource. The HERE HD Live Map is the deeply detailed machine-to-machine map that autonomous vehicles need to perform tasks normally provided by human memory and experience. The map enables autonomous vehicles to understand their position on the road in a measure of centimeters, rather than yards. What’s more, it contains information about the road ahead – so a vehicle can see and take action based on data beyond the range of its own sensors.

The second requirement is a powerful GPU that does the job of human decision-making. That GPU must pack enough processing power to ingest all the information from the road and render smart decisions that keep drivers and passengers comfortable and safe in real-time. NVIDIA is building these processors, and they’re bringing them to market without adding thousands and thousands of dollars to the cost of your vehicle.

We’re proud to be building autonomous driving solutions with our partners at NVIDIA. We're excited to help implement the tasks you’ll want in a highly autonomous vehicle.