Start > Autonomes Fahren > Autonomes Fahren: Was macht NVIDIA?

Autonomes Fahren: Was macht NVIDIA?

Heute schauen wir mal hinter die Kulissen von NVIDIA, einem der Schlüsselexperten für das autonome Fahren. Sehr spannend!

NVIDIA, which currently has a massive lead in this space both on the deep learning side and inference side of the AI (artificial intelligence) solution just stepped up with their Hyperion Kit and I think it is another game changer.
NVIDIA Hyperion
At the heart of the NVIDIA Hyperion Kit is the Drive AGX DevKit which is the brains of the solution. This is basically a small supercomputer with multiple 8-core “Carmel” CPUs, a Deep Learning Accelerator, a Volta-class GPU, programmable vision accelerator, stereo and optical flow engine, an image signal processor and video encoder/decoder. The kit adds to this 7 external and 1 internal camera (for driver monitoring), 8 radars, and an optional Lidar component. This provides a comprehensive vision package that should perform in all reasonable weather and light situations and—since the NVIDIA solution is fully trained—provide a near plug-and-play experience once the solution is properly installed and introduced to the car.
Other than installing the computer and sensors, assuring the car has the necessary control infrastructure for self-driving is also necessary—suggesting cars that currently use computer-aided breaking and drive-by-wire systems would be the easiest to upgrade.
The benefits of this effort are largely time to market. The kit buyer doesn’t have to qualify the sensors, can get help from NVIDIA as to sensor placement, and doesn’t have to spend time rethinking parts of this solution that are cooked. The result is they have a higher base to start from and thus can cut years out of what would likely otherwise be a 3 to 5-year development process. While not as useful for large car makers—because they are already well down this path—for smaller manufacturers this could make the difference between having a viable solution or not.
Having a larger number of cars capable of at least Level 4 self-driving could massively accelerate the adaption of this technology and help bring needed regulations on line more quickly. With the result that a viable self-driving solution at scale across multiple vendors could be more effectively brought into this decade.
Eventually I think this progression will result in an NVIDIA car, not for sale to consumers, but as a rolling kit that will allow those that are still late at the end of the decade to catch up. You see, we still need have a full concept of what changes to existing car design concepts need to be made to fully-optimize the switch from human to computer and NVIDIA will—thanks to their very early start in this market—have the best idea how to re-conceptualize this. Things like where the passengers should sit, whether the vehicle should have windows or just displays, and technology placement—particularly on electric drive vehicles which are, largely thanks to Tesla, again coming into market in volume.

Keine News mehr verpassen! Jetzt den  Newsletter abonnieren .
Menü schließen