Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta Videos (and questions for FSD Beta drivers)

This site may earn commission on affiliate links.
Seemed like the Waymo waited for an eternity to merge in.

It was calling its mommy for permission :p

I travel a road like this every time I go to the grocery, and always get in the left lane very early, as there is very little warning that if you are in the right lane you must turn right. Human drivers unfamiliar with this get caught all the time, and usually always annoy other drivers when they stop and signal to get back in the left lane, as the traffic can be heavy. I think the smart and courteous thing to do in such a situation, is to continue the right turn, and look for a good place to turn around, or u-turn if such is available.

You may be familiar with the road, and that's why you're not caught in the right lane. I wouldn't blame the other drivers for "inconveniencing" traffic. It could be that the road is just poorly designed / confusing.
 
Last edited:
  • Love
Reactions: Nrkl
#FSDBeta 6 version 2020.44.15.4 Part 2 of a demonstration of the colors and descriptions of the objects that are used in the FSD visualizations....
THX, just a slight addition on Part 1. You should have added the the solid DOT indicates the perceived rear of the vehicle.

Also glad to see the Pink and Purple explained. They are hard to see the difference and I didn't realize it was actually 2 different indicators.
 
  • Like
Reactions: byeLT4
Hard to describe, but I'll say I could see better than the gopro, but not as well as you can see from the TeslaCam footage :)
Got any dark tunnels/underpasses to test with your stroller/garbage can? ;) Here's a clip from a video diplomat33 shared a while back, and the camera basically shows completely black tunnel until a few meters away:
tunnel stroller.jpg

But probably what worry me the most are cases where cameras are likely to fail.

- Cameras being blinded
- Cameras not seeing quick enough a pedestrian with dark clothes or dark animal at night
- Cameras not seeing quick enough a pedestrian hiding in a shadow like my dark tunnel example.
- Cameras not recognizing sharp glass on the road because glass is see-through.

It would be interesting to see if TeslaCam always has better view of the dark or something dynamically adjusts the exposure based on the conditions, so a tunnel is tricky in that cameras might have exposure settings adjusted for bright sunlight. I suppose I could try testing this out myself even without FSD beta! :p
 
  • Funny
Reactions: DirtyT3sla
#FSDBeta 6 version 2020.44.15.4
This is a demonstration of three different lane merges one onto an Interstate and two onto local highways.

first merge onto the freeway took a little too long for my comfort to get up to proper speed.

1:37 - I wonder if the double dotted lines confused the system and made it think it couldn’t change lanes? Not really acceptable for the car to be darting from the right turn only lane into the straight lane at intersections.

2:20 - the guy who passed was an asshole, IMO, but that’s a situation that Tesla needs to program better. Either speed up to get in ahead or slow down to slot in behind. Car might have been fine without the over-ride, but I don’t blame the driver for taking over.

edit - as was pointed out, there was a yield sign for the Tesla. The car effed up there.
 
Last edited:
...2:20 - the guy who passed was an asshole, IMO, but that’s a situation that Tesla needs to program better. Either speed up to get in ahead or slow down to slot in behind. Car might have been fine without the over-ride, but I don’t blame the driver for taking over.
The Tesla ran a Yield sign. It was not a merge lane.
 
It would be interesting to see if TeslaCam always has better view of the dark or something dynamically adjusts the exposure based on the conditions
Looks like it's more of direct glare washing out the rest of the view. Based on Dirty Tesla and AIDRIVR night time videos, I would expect Autopilot at night to render the pedestrian in the tunnel without issue even wearing black:

full glare.jpg


Kinda neat to see how clear the view of the tunnel becomes even when the sun is partially blocked for 1 frame out of 36 (28 milliseconds):
partial glare.jpg


And as usual, these TeslaCam images are lower quality compared to the ones fed into the neural network.
 
  • Informative
Reactions: pilotSteve
Hard to describe, but I'll say I could see better than the gopro, but not as well as you can see from the TeslaCam footage :)
Yes - at night, esp. when its raining, AP sees lane marking much better than I can. I think something everyone who wears glasses can probably attest to ...

ps : Check the web for night time / low light image enhancements that NN can do. Its a real "eye opener". I've linked it previously.

Like this ...

Low Light Image Enhancement with CNN | Kaggle

140415-ASUS-PixelMaster.jpg
 
I really don’t think testers should be slowing down to see if it works. If you are controlling the speed that is not FSD in my opinion. If it doesn’t work without u slowing down it doesn’t work no matter how cut it.

100% agree. Even the current state of AP now doesn’t slow down enough for curves. I find myself constantly disengaging on certain curves because it’s getting way too close to the centerline. It’s was fine back on 2020.28.5.
 
  • Like
Reactions: HighZ