Self-driving cars would be nowhere without HD maps
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Annelise Capossela/Axios
Self-driving vehicles may be loaded with sensors and artificial intelligence, but they're limited without a really good map.
Why it matters: High-definition maps are critical to the safe, wide-scale deployment of autonomous vehicles. More accurate than satellite-based GPS, they provide richly detailed models of the operating environment and important context to help AVs avoid mistakes.
Driving the news: A Tesla owner tweeted a video clip recently showing how his car's Autopilot system mistook a low-hanging moon for a yellow traffic light and kept telling the car to slow down.
- While Tesla did not publicly address the reasons for the error, industry experts suggest Tesla's camera-based system was lacking important context.
- "Even though a traffic light and the moon may resemble each other, a self-driving system should use a combination of contextual cues — including spatial, temporal and prior knowledge — to tell them apart," Deva Ramanan, principal scientist at self-driving tech competitor Argo AI explains in a blog post.
- An HD map — along with redundant sensors like radar and lidar — can provide that missing context, Gartner Group mobility analyst Michael Ramsey tells Axios.
- "It would know where there are traffic lights. The moon is not a yellow light because there are no traffic lights in this area," he said.
The big picture: Mapping is having a moment. Digital maps are getting more sophisticated, with breakthroughs enabling real-time navigation details for pedestrians, 3D geolocation for drones and augmented reality for gaming.
- For autonomous vehicles, HD maps do more than just provide a high-def view of the world — they also enable a self-driving car to know precisely where it is, down to a few centimeters.
Between the lines: Most AV developers get to know a test city the same way any new resident does: by driving around.
- They spend a few weeks manually driving their test cars in complex urban environments, collecting sensor data and annotating everything about the streetscape, from signs and lane markings to crosswalks and speed limits.
- This allows a test vehicle to later compare what it observes in real-time with the detailed 3D map, and decide how to react.
- Developers can even program insights about local driving behaviors into their digital maps. AVs can be instructed to drive below the speed limit on a certain stretch, for example, or to stop beyond the line for better visibility at a challenging intersection.
- Developers can repeat that map-making process city by city.
The intrigue: Intel-owned Mobileye, which makes technology for assisted-driving systems, has a unique, crowd-sourced approach to mapping that experts say could one day provide an advantage.
- Its camera-based software chips are already installed on 88 million cars worldwide, and through agreements with six global automakers — including Nissan, Volkswagen and BMW — many of those cars collect and share data about their environment as they are being driven, allowing Mobileye to continually update its maps.
- With better mapping, Mobileye will be able to more rapidly scale AVs across multiple cities, CEO Amnon Shashua tells Axios.
Yes, but: Mobileye, which recently started testing AVs in New York, still has a long way to go with its self-driving technology, notes Ramsey.
The bottom line: With HD maps, though, at least they know where they're headed.
