If I understand correctly Tesla is/has removed LIDAR and uses computer vision for most/all of their self driving.

https://bdtechtalks.com/2021/06/28/tesla-computer-vision-aut...

I don't think Tesla ever used LIDAR and the article confirms they don't think they will need to. I believe they removed ultrasonic sensors though, maybe that's what you're thinking of.

Yes, perhaps the single-most controversial decision Tesla has made regarding FSD.

Everyone else uses LIDAR in some form. Tesla's cameras can and have been fooled on many occasions.

Yet they stockpile them:

https://www.theverge.com/2024/5/7/24151497/tesla-lidar-lumin...

To be fair, $2MM of LIDAR units seems more like a R&D purchase than a stockpile.

$2.1m?! They must have bought two!

2,100 - according to article

Yeah okay, but that doesn't mean _cameras_ are bad (which, to be fair, they are in Teslas case), it means the algorithms feeding on them are.

It means the cameras can be fooled by things LIDAR cannot be. Such as smoke, glare, reflections, optical illusions/mirage, etc.

If the algorithms are fed with incorrect data, they will produce incorrect results - such as driving full-speed into a parked, white colored, semi-truck.

And lidar can't tell the difference between a plastic bag and a rock, what's your point?

One can (and has been) fooled into thinking there is no object in the path - the other might be extra sensitive to any object in it's path.

I'll let you stew on that one for a minute...

> I'll let you stew on that one for a minute...

Then that means the vision processing isn't far along yet to be viable for a car. There is no fundamental reason why it couldn't work though. With either stereoscopic vision or more temporal processing you could obviously detect when things are only painted on a wall surface, with both there really is no excuse to still fail except limited processing power.

And this is likely why Tesla's FSD is... not very good.