Autonomous driving, NVIDIA brings real and simulated worlds closer with Omniverse Replicator

0
158

If Facebook Meta has the metaverse, NVIDIA is not; less with its Omniverse , the true protagonist of the keynote of CEO Jensen Huang during GTC 21. And there are not only avatars in the guise of digital assistants 2.0, but also advanced solutions for the & # 39; automotive and autonomous driving . As explained by the same company & agrave; American, & quot; the gap between reality & agrave; and simulation & egrave; just been reduced & quot ;.

OMNIVERSE REPLICATOR

It is called Omniverse Replicator , and it is; a synthetic data generator to train artificial intelligence networks and improve autonomous driving systems. The simulation tool made by NVIDIA and based on Omniverse * & egrave; named DRIVE Sim : the (synthetic) data generated by it is used to train the neural networks “ which constitute the perception systems in autonomous vehicles ''.

In turn, these networks are composed of a model algorithmic and the data used to train the model itself . If on the former – the algorithms – the work is; constant and intense, on the second – the data – greater difficulties occur; due to their scarcity; in the real world – either for the lack of time , or for the high costs . Hence the importance of developing synthetic data that go alongside (and enrich) the real ones (usually collected through sensors mounted on the cars under test).

* & quot; Platform open and easily extensible built for virtual collaboration and physically accurate simulation in real time & quot; – NVIDIA

Why & eacute; the system is working properly, however, & egrave; the simulated data must be the most & ugrave; possible consistent with the real ones . There are therefore two types of differences that can occur between simulated and real world:

  • differences (at pixel level) between image simulated and real
  • differences between the context simulated and the real one (for example: the dirty road in the real world must also be replicated in the simulated one, as well as traffic, pedestrians and various interactions)

Here is where Omniverse Replicator comes into play:

DRIVE Sim leverages Omniverse's RTX path-tracing renderer to generate sensor data for cameras, radar, lidar, and ultrasonic sensors. Real world effects are captured in the sensor data, including phenomena such as LED flickering, motion blur […].


DRIVE Sim's task is; to reduce the pi & ugrave; the gap between the real world and the simulated world is possible by leveraging Omniverse “ to connect to a “ wide range of tools for creating content ''. Thus, Omniverse Replicator & quot; organizes data for quick scene manipulation using a feature; called & quot; domain randomization & quot; & quot ;.

NVIDIA already does; use of this solution for the generation of synthetic (simulated) data which in turn reproduced millions of images for autonomous driving training , integrating them with variables such as weather conditions, traffic, road debris and pedestrians. And the level of precision and detail with which this data simulates a scene is; higher than in the real world. For example, information not visible to the human eye is for Omniverse Replicator .

The best of OnePlus? OnePlus 9 Pro, on offer today by atpservicepomezia for 758 euros or from Amazon for 839 euros .