I am a fan of and impressed with autopilot. It’s a great addition to my commute. It’s also the only system I’ve used that is beyond basic cruise control. I’m always impressed and sometimes amused by the images of other cars (and cans/cones) on the screen. This morning it decided a trailer spool of hose was a motorcycle! I wondered if cars with more than vision are more accurate in this? Any thoughts?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
UI displayed info ≠ FSD computer operation. The UI is VERY limited on what it displays and a spool of hose is NOT one of them. The list of what it can't display like RR crossings, trains, school buses, firetrucks, trees, lamp poles, boats on trailers, scooters, wheel chairs.......is to near ∞.
Can YOU identify it with your vision [rhetorical since you did] ? Adding any other info beyond vision would not change your interpretation of what it is. Not saying lidar or radar can't be a plus but your example is completely irrelevant at demonstrating a use case scenario.
Can YOU identify it with your vision [rhetorical since you did] ? Adding any other info beyond vision would not change your interpretation of what it is. Not saying lidar or radar can't be a plus but your example is completely irrelevant at demonstrating a use case scenario.
RTPEV
Active Member
Still, it's humorous that it apparently mistook the spool for a motorcycle wheel....that's a BIG motorcycle!
I don't have a picture to share, but another interesting one is how it interprets Amish buggies (of which there are a lot in my in-laws area). It seems to recognize the horse as a dog, with the person riding behind in the buggy as just a regular person walking behind the dog, although it flip flops between that and a truck (can't quite make up its "mind").
I don't have a picture to share, but another interesting one is how it interprets Amish buggies (of which there are a lot in my in-laws area). It seems to recognize the horse as a dog, with the person riding behind in the buggy as just a regular person walking behind the dog, although it flip flops between that and a truck (can't quite make up its "mind").
EVRider-FL
Active Member
It doesn’t really matter as long as the car detects the object and avoids crashing into it. 
Kimmi
Member
My understanding is that the FSD computer reads data from the cameras and creates a volumetric view of the world, called the occupancy network. It's just blocks of shapes that the car knows are there. There is a transformer subroutine that looks at those shapes and gives us the visualization of what it thinks the shape is. That subroutine is cosmetic for our benefit on-screen and doesn't change driving. The NNs just care about drivable space vs non-drivable space.
The point is that the workbench is not something the subroutine knows about, but its shape was enough of a match with what a truck looks like so that's what the visualization showed. It wouldn't drive into it.
The point is that the workbench is not something the subroutine knows about, but its shape was enough of a match with what a truck looks like so that's what the visualization showed. It wouldn't drive into it.
Yes, I completely understand that the programming has a limited range of display outputs. I see it switch between a semi and a pickup with a trailer and understand it. The leap from large spool of hose to motorcycle is more puzzling, and I thought enjoyably humorous. Please don’t let this forum because another soulless section of the internet devoid of any thought beyond ELON=GOD!!! Or ALL COMPUTERS BAD. The car showing a semi trailer inside my garage is exactly the kind of thing a person who would put emissions mode in a vehicle would also enjoy. This post is for them.
Similar threads
