No, we'll be coming down on 75.

 

Question for those that use FSD fairly regularly - considering that FSD gets better with training (actually driving), can it be trained to slow down "below" the posted speed simply by "reducing" (using the scroll wheel to reduce the speed while it's in FSD) your driving speed when going over the speed bump?

So the next time you're driving over the exact same speed bump (in my neighborhood as an example), would it go slower than the posted speed limit?

I'm assuming this scenario can also be applied to going around a pothole or slowing down over very uneven rail road tracks

 

i think training is overall for every FSD instance out there, not for your particular vehicle and your own way of driving.

Something like speed bumps, they need to program in somehow.

 

FSD in your car does not get better as you drive, only a new version brings improvements. FSD gets better from watching you drive via training on the mother ship.

Xpeng is doing a thing that seems familiar

 

LLMs can are often multi-modal. You can combine vision and LiDAR as part of model training. They are not separate data sets - can be passed in at the same for training with time synchronisation using semantic data models.

The real reason is cost to do so, cameras are cheap. LiDAR is not.

 

LiDAR is a crutch, it gives the system extremely precise distance telemetry which compensates for poor vision interpretation. LiDAR does not replace vision cameras or there wouldn't be any, and it can't see colour. This leaves sensor fusion confusion.

 

people might not realize the ‘only’ cameras on a tesla are still WAY better than human eyes, and there’s 8 of them.

integrating lidar and camera data is apparently very challenging.

driving the y last night i was curious to see what it could see at night and it identified pretty much everything out there as far as i could tell. amazing.

 

full self driving upgrade (14.1) seems to be insanely amazing. can’t wait to get to try it soon.

 

To the camera at night point, you only have to open up a modern phone camera at night, in seemingly near dark, to see that there is still an insane amount of detail that can be retrieved. The cameras work their low light magic via longer exposure times and other processing, so not necessarily ideal for a moving vehicle, but that doesn't change the fact that cameras can pick up detail we can't in low light.

 

That FSD 14 vid is incredible.

 

Very cool! 👍

 

What is so incredible about Omar's video?

 

Not the vid necessarily. The improved parking options in v14 and park at a charger is what I found cool.

 

Meh, not impressive to me. This is kinda good.


Until Cybercab is driving around collecting fares I'm not going to give Tesla credit for solving autonomy.

 

When you drive an OG Tesla that just has TACC, any version of FSD is cool.