Quote:
Originally Posted by bavarianride
Quote:
Originally Posted by pikcachu
But I've never seen any other vehicle be able to stop and turn on regular streets like Tesla does :
Maybe it's cause others only work on freeways… Waymo does work on streets but it has 500lbs of boxes/stuff inside/on top of the vehicle. What other me normal looking car can do "self driving" like Tesla?
Also just the amount of data Tesla can collect should be 100s times more than any other manufacturer, so seems like they should be able to handle edge cases much better
|
A hint provided in your post is "amount of data Tesla can collect", so basically it is edge inference with centralized learning, and we know how well that has been working.
One option may be to move learning onto the car but the power profile(plus extra weight) will drain the battery and range like crazy.
So the practical solution right now is similar to Cruise, which is fixed hwy topology(no need to relearn) and incremental real-time learn + infer. I think that is quite reasonable with current system capabilities.
But obviously Tesla cannot tell u that or else that cannot overcharge $10k for their L2 ADAS.
|
You aren't going to do edge learning.
Like you said the weight and power are.prohibative.
Also adaptive algorithms in a realtime are a bad idea.
Cruise, BMW and others have a more conservative approach that is less likely to kill or injure.
MobilEye has a fully redundant system of processor and system input. Continental does also. There has been some consolidation in the area and I expect more in the next couple of years.
MobilEye is Intel.
Nvidia has a platform.
Cruise is owned by GM.
Continental is contracting for others to do chip development.
We aren't to the level of FSD in regular passenger cars. It's still specialized with big boxes on top of cars and very few licenses granted in very specific areas for FSD. The leaders are Cruise and Waymo.
https://www.google.com/amp/s/www.cnb...less-cars.html