Last night I watched this video by Teslas AI dude:
its basically about the challenges, and approaches tesla is taking to FSD. Its REALLY wworth watching. I'm a computer programmer, and have dabbled with neural net stuff, so I was especially interested, but its interesting for everyone and I thoroughly recomend every investor watches it. If I was not married and living and working from home in rural england, I'd be applying to work at tesla right now...
Anyway, the short summary:
Recognizing stop signs is HARD, because although the sign is simple, the circumstances for the edge cases are insane. Some stop signs are lit up at night, making themn *harder* to see for cameras. Some are on buildings, some are obscured by trees, some are on a stick held by a construction worker. Some are on barriers that lift. Some have writing underneath them that modifies the intention oif the sign.
The real big one for me is... a construction worker holds a stop sign, and its valid, when he lowers it (but you can STILL see it), its no longer valid, and the car can proceed. These edge cases are really hard to detect, but tesla is working on it and has a massive advantage in collecting data on it.
Basically what I'm saying is that the edge cases than joe-public assumes are impossible (stop sign not normally there, but a dude was holding one on a stick) are not only on teslas radar, they are wworking through them RIGHT NOW. And like Elons approach to spacex (dont built the rocket, build the system that builds the rocket), Tesla are building an AUTOMATEd system to handle cases like this, identify them, categorise them, analyze them and update the fleet with them.
Shorter version:
Waymo are ****ed. Everyone else is ****ed, Uber is ****ed. Teslas FSD system is light years ahead already, without even stretching the ability of the HW3 chip.
Watch the video, it will make you much more confident about the possibility of real FSD. I was flabbergasted. And automatic-distance estimation per pixel just using cameras...LOL. Lidar is such a dead end.