Musk is again going against common industry wisdom. He proved naysayers wrong many times with Tesla and SpaceX, but will this be the case here? Humans are able to drive just using two eyes, but do we really have good enough AI to build self-driving just based on cameras, ultrasound and radar?
- Humans are able to drive purely through subconscious information processing, leaving your conscious mind free to think of the past present or future. You can do this continuously unless there’s a sudden action that triggers your conscious mind to respond to sudden conditions.
Lidar is just a modeling technique to understand the environment around you. Is it better than how our subconscious works? That’s a good question if EO/IR image processing and ultrasound can model this equally as well. I’m skeptical of Elon on this one given my naive armchair analysis of this. But generally the wisdom of the masses is smarter than the Wisdom of the few. But I’m nowhere near as smart as this guy either.
- I agree.... it’s redundant info in a sense and introduces more edge cases normal humans don’t have
- Microsoft JKWj76Hard to argue he is wrong here. He said "relying" not just "using".
One thing he is right: It does feel wrong to have a FSD stack that has such a sucky vision stack that has to be helped by Lidar. That implicitly means that this FSD stack would simply not work with vision alone when Lidar is not providing critical info about the scene.
Now, if Lidar is not critical to the system (i.e. self-driving works just as well through vision alone in the corner cases where Lidar doesn't work) then the next question becomes whether it's cost effective to include with every car, if it's just an expensive add-on that only provides some minor incremental value.