It probably can't be 'fixed' - vision only + 'end-to-end NN'. I don't believe you can get it to be really reliable with that setup, at least not in this decade.

It probably can't be 'fixed' - vision only + 'end-to-end NN'. I don't believe you can get it to be really reliable with that setup, at least not in this decade. It's a very good setup for awesome demo of the technology and vague promises in the future. Just like in 2016 (every Tesla will become a Robotaxi with an OTA update) and 2019 (we will have 1 million Robotaxis on the road in 2020). Another example - Tesla official account highlighting that FSD is 'taking pedestrian gestures into account': https://nitter.poast.org/Tesla/status/1776021631346835491#m https://nitter.poast.org/Tesla/status/1776021631346835491#m Is it a lie or not ? Even the usual influencers are saying it's not taking pedestrian gestures into account, based on their tests. But, the system is an NN - so it's impossible to prove it - just like it's impossible to prove that it does take it into account. Hell, it takes everything it sees on the camera into account !