Tesla Full Self-Driving to feature ‘end-to-end AI’ with groundbreaking v12 release

Update: para 5 updated for accuracy. 

Tesla’s Full Self-Driving suite will be vastly different from its current buildout when the automaker eventually releases the now-groundbreaking v12 build, CEO Elon Musk said. Tesla plans to utilize end-to-end AI with the v12 release, Musk said.

Upon the imminent release of v11.4 to employees this week, Musk said the new version will feature “so many major improvements” and that it should be v12. However, v12 features a new strategy with “end-to-end AI,” Musk said, which is a drastic difference compared to the current state of FSD.

Arguably, v11.4 should be v12.0, as there are so many major improvements.

v12 is reserved for when FSD is end-to-end AI, from images in to steering, brakes & acceleration out.

— Elon Musk (@elonmusk) May 7, 2023

End-to-End AI is a strategy where the model, in this case, FSD, learns all steps between the initial input phase and the output result.

Tesla has used neural networks with its FSD suite to help with learning, and it improves with every mile driven. The millions of miles traveled with FSD by owners have helped make the system more robust and improve over time in terms of behavior, but this would be a major change from what Tesla has normally done.

Today’s vehicles utilize cameras for inputs, which output a three-dimensional representation of the environment that surrounds the car. It then uses traditional path algorithms for outputs like steering, acceleration, braking, and other driving behaviors.

End-to-end would only utilize the cameras on a Tesla vehicle to input the neural network, while the output would be the steering, acceleration, braking, and other driving behaviors. The entire FSD suite would be coded in a single neural network and can be learned by relying on driving behaviors input by humans.

End-to-end models require less work to build component-based systems, which simplifies the process and could help Tesla jump even further forward with its development of self-driving software. It also is much more optimized because it uses only a single neural network.

There are also drawbacks, like a lack of predictability and more difficulty diagnosing problems.

As for v11.4.1, Musk said it is “a major improvement” with some driving behaviors, as owners expressed some concern with basic maneuvers.

I’d love to hear from you! If you have any comments, concerns, or questions, please email me at [email protected]. You can also reach me on Twitter @KlenderJoey, or if you have news tips, you can email us at [email protected].

The post Tesla Full Self-Driving to feature ‘end-to-end AI’ with groundbreaking v12 release appeared first on TESLARATI.