• 0 Posts
  • 6 Comments
Joined 2 years ago
cake
Cake day: June 23rd, 2023

help-circle




  • To be fair, the Tesla vision system has 3 cameras facing forward. One in the center above the front bumper grille and two behind the rear view mirror. Those two provide some level of stereoscopic vision to help judge distances.

    But yeah, the lack of other sensors is a huge issue. Anything from bug splatter to mud to snow etc. can easily obscure one or more cameras and render the whole vision system unreliable.

    We also process light differently than cameras do

    To expand on this a little further, human vision has also developed the ability to filter out unnecessary information in order to avoid overloading the brain. When tracking moving objects the eyes mostly send deltas of the movement to the brain. Computers, however, are the exact opposite. The cameras essentially send a series of still images, and it’s up to the computer to compare them to look for any movement.



  • On top of this they are also highly compartmentalized, which sounds like a nightmare to me. A few years ago we ran into a bug with one of their AWS services (I forget the specifics at this point, but it was with a network load balancer). It was a rather novel bug, but easy enough to reproduce, so I was able to give AWS support a working example. They quickly confirmed the bug and said it would be fixed as soon as possible.

    Since it was an odd bug I asked AWS support if they could provide a high level description of the fix once it was implemented. It was then that they told me how their teams were so highly siloed and couldn’t really share details like that with each other. The support rep I worked with wouldn’t even know if the bug was fixed by a load balancer team, a more generic networking team, or some other team. They would only know the bug was fixed.