The name of You’re here often associated with autonomous driving, although the system as it is designed for now only falls under level 2 on the SAE scale, that is, the one that always requires driver assistance. But, as we’ve known for some time now, Tesla is preparing the so-called Fully autonomous drivingand to do this, he implemented a new software system and also a kind of artificial intelligence that learns to drive the car.
But the software alone cannot learn, it needs to have a detailed description of its environment, thanks to thousands of examples of similar situations which allow it to identify obstacles, other cars, people or animals, or just the side of the road. To do this, there are several Tesla employees who they take care of the “labeling” everything that appears in the videos taken by the on-board cameras, but this work is not enough.
1/3 Some panoptic segmentation eye candy 🌈 from a new project we’re talking about. These are too raw to work in the car, but power auto labelers. Collaboration between data labeling of large dataset (over 100,000k), clean, diverse, multicam+video and engineers who train the models pic.twitter.com/RTERAxyRO0
Andrej Karpathy (@karpathy) November 30, 2021
Also taking advantage of the Dojo Super ComputersTesla can pass the images to an algorithm, which takes care of independently label what it sees. At the moment, the project is only in its infancy, but the head of division Andrej Karpathy has shared on the Internet images that look very promising. The AI recognizes vehicles, pedestrians, motorcyclists, various structures and road signs.
If the preliminary results hold up over the long term, this trick up its sleeve would add to one that Tesla has been using for a long time: the number of cars already sold. Right now, the California house has about a million cars on the road, and they all provide data and video to parent company, even those who do not officially have assisted driving at the highest level. All the collected data is too heavy for the on-board computers, but when passed to the super computers, it can generate a huge plus in the teaching of the road. For this quantitative and qualitative combination, Tesla is still considered to have a clear advantage over the competition in the field of autonomous driving.