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.
“You want to give it as much leeway as possible…to give it an opportunity to correct itself.”

Why do people believe this? There is zero benefit to exposing it to challenging situations and “seeing what it does” without intervening. What downside is there to intervening early and often? None that I know of!

It’s funny how he thinks he is out just $500-$600.

That is a ~$2k collision with 20” wheels.
View attachment 721395
Just catching up with the thread. I've been saying the same thing and the only answers I can come up with is that, spectators want to see the outcome and secondly some people believe they are helping by letting the system fail before reporting or disengaging which is clearly not true.
 
To whoever gets into an accident on FSD, it's like falling in your own driveway - it's your own asphalt.

Don't blame a beta (really, alpha) software to save you. We already know in several ways it's trying to kill you. Certainly don't risk having kids or others in the car until you can easily predict when it's likely to screw up. Kind of a Russian roulette except it's a hand grenade instead of a gun (others may be hurt) but you are allowed to have a hand on the pin - but you are still choosing to play.
 
... Certainly don't risk having kids or others in the car until you can easily predict when it's likely to screw up. ...
We each have our different risk levels we tolerate. I plan to have my children in the car. I also plan to take over quickly if it is not doing what I expect. It also matters that my only commute is taking children to school and picking them up.
 
  • Like
Reactions: rxlawdude
Why on earth were they letting it continue? These testers are nuts. There's zero value in letting it do the wrong thing for any period of time at all. Except maybe for clicks. Mission Accomplished, I guess? Should have intervened around 5:45 or 5:46 as it deviated from the correct path.

~4:30
"Kind of feels like being driven around with an Uber. Occasionally my Uber makes some mistakes."

"Right up on par with an average Uber driver."

"Honestly!"
The worst part is:

"I really apologize to Tesla...." 🤦🏽‍♂️
 
A Progression of Forward Facing Unprotected Left Turns on #FSDBeta v8.2 - v10.2 - Check it out.

Sometimes we miss incremental improvements in each version 'point' release but Chuck's consistent approach to testing his routes and an excellent compilation video like this give us a much better perspective on FSD beta progress over time. Thanks Chuck.
 
I at basic level know how NN image recognition works. What is the plan to avoid things that are not recognized?

@mark95476 I see that you disagreed with this. Why? This is a genuine question.

NN classifies objects. At the moment NN in the FSD can recognise e.g. other cars, cyclists,pedestrians and knows that it can’t collide with them. It also reconises e.g. trash cans, curbs and traffic cones and doesn’t collide with them.

But at the moment it obviously doesn’t recognise everything. I’m curious what is the plan to do if it sees on the road something that it can’t classify. Is the plan to always stop in that situation or something else. If the amount of unrecognised things is small, this can work. But if the amount of unrecognised things on the road is large, this leads to many false braking.

Currently apparently if fsd doesn’t recognise something, it doesn’t care about it.
 
  • Like
Reactions: Phlier
@mark95476 I see that you disagreed with this. Why? This is a genuine question.

NN classifies objects. At the moment NN in the FSD can recognise e.g. other cars, cyclists,pedestrians and knows that it can’t collide with them. It also reconises e.g. trash cans, curbs and traffic cones and doesn’t collide with them.

But at the moment it obviously doesn’t recognise everything. I’m curious what is the plan to do if it sees on the road something that it can’t classify. Is the plan to always stop in that situation or something else. If the amount of unrecognised things is small, this can work. But if the amount of unrecognised things on the road is large, this leads to many false braking.

Currently apparently if fsd doesn’t recognise something, it doesn’t care about it.
Musk says that FSD Beta can determine the height of a pixel above the roadway without classifying it. His example is if a UFO lands in the street, don't hit it.
 
  • Like
Reactions: Phlier and Matias
Apparently it is not easy since they're not yet doing it.

Nothing Tesla has achieved with vision has been easy. I'm constantly blown away by what's going on behind the scenes. Just a few examples of challenges many thought (myself included) Tesla would have trouble with or never solve:

Direct sunlight
Heavy rain
Not enough cameras
Shadows
Train tracks
List goes on...

Predicting the height of road objects is just another challenge Tesla will overcome with vision (hopefully). At this point, I have little understanding as to how Tesla is achieving their results, as it's basically magic.
 
Nothing Tesla has achieved with vision has been easy. I'm constantly blown away by what's going on behind the scenes. Just a few examples of challenges many thought (myself included) Tesla would have trouble with or never solve:

Direct sunlight
Heavy rain
Not enough cameras
Shadows
Train tracks
List goes on...

Predicting the height of road objects is just another challenge Tesla will overcome with vision (hopefully). At this point, I have little understanding as to how Tesla is achieving their results, as it's basically magic.
It does seem like cameras are the way to go. Amazon’s Astro robot uses vision rather than lidar as mentioned at 1:45 in this video.
 
  • Like
Reactions: Phlier and EVNow
Musk says that FSD Beta can determine the height of a pixel above the roadway without classifying it. His example is if a UFO lands in the street, don't hit it.
Yes, he has said this. He may mean that if they turn on that capability (voxels?) then it can do this? But it does not appear to be turned on. From various videos it appears that there are many objects that are just not recognized and are ignored for now. (I don’t have another explanation for it trying to run into a post.) Fortunately have not yet had personal experience with this phenomenon of unidentifiable objects. I definitely have zero confidence that it can identify unusual objects in the middle of a road.

I think there are enough people with the beta now it should be possible to figure out what it can identify and what it simply ignores.
 
  • Like
Reactions: Phlier and Matias
Yes, he has said this. He may mean that if they turn on that capability (voxels?) then it can do this? But it does not appear to be turned on. From various videos it appears that there are many objects that are just not recognized and are ignored for now. (I don’t have another explanation for it trying to run into a post.) Fortunately have not yet had personal experience with this phenomenon of unidentifiable objects. I definitely have zero confidence that it can identify unusual objects in the middle of a road.

I think there are enough people with the beta now it should be possible to figure out what it can identify and what it simply ignores.

What about this wonky tree maneuver? Or the monorail? I think it's activated, but it's not very good or only maneuvers when confidence is high.

 
What about this wonky tree maneuver? Or the monorail? I think it's activated, but it's not very good or only maneuvers when confidence is high.
It looks like the NN extrapolated the double yellow line through the tree causing the car to go around it to correctly position itself in the lane.
1634484618627.png