Automotive

NVIDIA's Drive Map to chart 500,000 km of roads for autonomous cars

View 2 Images
The LiDAR layer on NVIDIA's Drive Map platform
NVIDIA
The radar layer on NVIDIA's Drive Map platform
NVIDIA
The LiDAR layer on NVIDIA's Drive Map platform
NVIDIA

NVIDIA has been lending its expertise in computing to the world of autonomous vehicles for some time now, and a newly announced platform is designed improve the safety of these vehicles as they hit the streets. Called Drive Map, the technology will offer mapping of 500,000 km (310,00 miles) of roads around the world with centimeter-level accuracy, enabling self-driving vehicles to more easily find their way around.

Announced at the GTC developer conference this week, Drive Map is a mapping platform designed to accelerate the arrival of Level 3 and Level 4 autonomous vehicles by making it safer to deploy them. With more detailed awareness of their surroundings, the autonomous systems that steer these vehicles should be able to make smarter choices, improving safety for road users and pedestrians alike.

The platform is based on technology from startup DeepMap, which was acquired by NVIDIA last year, where highly detailed maps for autonomous vehicles have been in development over the past six years. This surveying technology will be packed into dedicated survey vehicles and work with a separate crowdsourced mapping engine for passenger vehicles. These will feed their combined data on roadways back to the cloud to build out a digital map of roads in North America, Europe and Asia.

The radar layer on NVIDIA's Drive Map platform
NVIDIA

The Drive Map will contain layers for camera, radar and LiDAR systems, with each serving a unique purpose for the autonomous systems at the helm. The AI controlling the vehicle might rely on the camera layer to track road markings, traffic lights and signs, the radar layer in poor lighting, and the LiDAR layer for a highly detailed picture at a 5-cm resolution.

According to NVIDIA, Drive Map will cover 500,000 km of roadway around the world by 2024, and will be continuously updated from there for use by the broader autonomous vehicle industry.

Source: NVIDIA

  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
4 comments
paul314
That's a good start, but will need lots more. New York City alone has well over 10,000 km of streets, and it isn't (geographically speaking) that bit a city.
VadimR
And how will you handle road debris, potholes, roadkill, ever changing construction, weathering conditions, repairs, etc. Seems like Tesla's method is a lot more reliable long term.
Jinpa
This project seems totally redundant to road information already digitized and public. Every public paved road already is in state and/or federal highway department files in digital form. It starts that way, because road design and modification is done digitally. Every feature of roadways, from the base grading and levels of the roads, bridges and tunnels to the final signage and all traffic controls is available. NVIDIA should have to explain why it isn't requesting and using that vast database of information. It ls mostly going to be a stupid waste of money. Shareholders and directors should do some serious holding of executives' feet to the fire before funding this boondoggle.
ljaques
With 4.09 million MILES of public roads in the US, this must include only a few large cities, eh Nvidia? DC to New York, maybe? Keep scanning and mapping, please.