Tesla’s FSD Beta v11.4 is making with its end-to-end architecture that inputs sensor data and outputs control signals. Elon Musk himself has called it a major improvement, so much so that he suggests it should be called v12.0 instead. This is because of the monumental shift from the previous modular architecture to the end-to-end architecture, which has completely pushed back the regulatory architecture.
The good news is that the software will be shipped to Tesla employees tomorrow, and as confidence grows, it will be progressively rolled out to a wider audience. However, testing this software is no easy feat, despite Tesla’s strong simulation and testing team. The reality of the road is vastly more complex than simulations and QA drivers can account for. It’s like the emergence of ChatGPT – even AI scientists cannot fully explain what happens in it.
Tesla is the first enterprise to attempt end-to-end large-scale engineering, with other experimental end-to-end projects coming from Nvidia and Horizon. With this comes the potential for pitfalls in engineering, but the benefits of end-to-end architecture are unmatched by the current modular architecture. The upper limit of the ability to deal with different scenarios is extremely high, but it needs to be tamed.
Despite Elon Musk’s claims that FSD is excellent, the reality is that the technology is very inconsistent across the United State, the State of Tesla’s FSD Beta v11.4 : Over-Promise & Under-Deliver?
As always, Tesla is pushing the boundaries of what’s possible with technology. It will be exciting to see how FSD Beta v11.4 performs on the road and what other advancements they have in store for us in the future.