Poll: How will youtubers rate FSD Beta 9?

[stopcrazypp]

Oh man it literally doesn't see the monorail columns/pillars

Does the visualization even have something for columns/pillars? You can't really tell if it can't "see it" as it seems a lot of what the perception engine sees is not shown in the visualization. Sometimes it is the path planning that goes wrong even if the perception is correct (for example the fence example shown further up thread where it renders the road correctly, but the path planning still tells it to go offroad into the fence). In this case, it seems to detect the solid white lines correctly, but path planning still tells it to cross them.

 

[Ptheven]

Why do you care so much about someone you clearly do not like?!?

Generally speaking, because they are being given access to things that people who paid before them are not, and that bothers me as it's just another way Tesla is screwing with customers.

I personally don't even care about auto steer on city streets, so it's not about me not having it. In fact it's very nearly worthless for me since most of my driving is on the highway, where we have seen no AP improvements of note in years.

 

[Dan D.]

Does the visualization even have something for columns/pillars? You can't really tell if it can't "see it" as it seems a lot of what the perception engine sees is not shown in the visualization. Sometimes it is the path planning that goes wrong even if the perception is correct (for example the fence example shown further up thread where it renders the road correctly, but the path planning still tells it to go offroad into the fence). In this case, it seems to detect the solid white lines correctly, but path planning still tells it to cross them.

I can't believe nobody has posted good videos of testing with large obstacles. Everyone manually stops just short of hitting things but we still don't know what happens if they don't stop.

I would set up some huge boxes with white sheets and see if it will run through them. Set up some boxes as 'parked cars' and push some people-dressed objects out from behind them. Don't want to hit boxes, use a bunch of big balloons covered with sheets then. Put these objects directly in the way, stack them like columns, drop them off a car that's 50ft in front of you.

If you're going to trust the car sometimes, but not know if you can trust the car other times, at least have some objective evidence either way. C'mon, have some fun out there.

 

[Daniel in SD]

I can't believe nobody has posted good videos of testing with large obstacles. Everyone manually stops just short of hitting things but we still don't know what happens if they don't stop.

I would set up some huge boxes with white sheets and see if it will run through them. Set up some boxes as 'parked cars' and push some people-dressed objects out from behind them. Don't want to hit boxes, use a bunch of big balloons covered with sheets then.

If you're going to trust the car sometimes, but not know if you can trust the car other times, at least have some objective evidence either way. C'mon, have some fun out there.

Do not trust the car until Tesla says you can. It needs to see pillars with a much higher reliability than an individual user can evaluate. People need to keep in mind that the human accident rate is about 1 per 150k miles and the fatality rate is about 1 per 100 million miles. The only objective evidence you might be able to get is that it's not good enough.

 

[stopcrazypp]

I can't believe nobody has posted good videos of testing with large obstacles. Everyone manually stops just short of hitting things but we still don't know what happens if they don't stop.

I would set up some huge boxes with white sheets and see if it will run through them. Set up some boxes as 'parked cars' and push some people-dressed objects out from behind them. Don't want to hit boxes, use a bunch of big balloons covered with sheets then.

If you're going to trust the car sometimes, but not know if you can trust the car other times, at least have some objective evidence either way. C'mon, have some fun out there.

The thing is such crude analogs may not fool the visual system and it may be able to tell it's not a real obstacle. And the more real looking analogs like used for AEB tests for example, are probably out of budget for most people. For this example, you would need to make something that looks like a concrete pillar, which a white sheet does not.

I also looked up the law given how the road is laid out makes it seem like it is designed to allow changing lanes between the monorail pillars (otherwise I don't see why there is no curb or other barriers added between the pillars). And guess what? It actually is legal to drive between the pillars to change lanes to another lane on the other side. So actually what the car did was legal. It's just without the pillars being visualized, I doubt anyone will feel safe letting the car do that.

Q: Is it illegal to change lanes under the monorail tracks that run down Fifth Avenue, splitting the road way in two so to speak?

A: No.

Is it illegal to change lanes under the Monorail tracks?

 

[Dan D.]

The thing is such crude analogs may not fool the visual system and it may be able to tell it's not a real obstacle. And the more real looking analogs like used for AEB tests for example, are probably out of budget for most people. For this example, you would need to make something that looks like a concrete pillar, which a white sheet does not.

I also looked up the law given how the road is laid out makes it seem like it is designed to allow changing lanes between the monorail pillars (otherwise I don't see why there is no curb or other barriers added between the pillars). And guess what? It actually is legal to drive between the pillars to change lanes to another lane on the other side. So actually what the car did was legal. It's just without the pillars being visualized, I doubt anyone will feel safe letting the car do that.


Is it illegal to change lanes under the Monorail tracks?

Stacked white boxes absolutely looks like a concrete monorail pillar. Same size, same shape, same color. 5ft tall or 50ft tall it should not hit them. A box 3'x3'x6' could contain a sleeping person. Does FSD Beta hit these things when they are in/adjacent to the road?

 

[Dan D.]

Do not trust the car until Tesla says you can. It needs to see pillars with a much higher reliability than an individual user can evaluate. People need to keep in mind that the human accident rate is about 1 per 150k miles and the fatality rate is about 1 per 100 million miles. The only objective evidence you might be able to get is that it's not good enough.

Sure. But DOES it see the pillars? It doesn't show them. All we know is that it seemed to veer towards them until the user manually swerved away. We don't actually know anything!

 

[Ptheven]

Stacked white boxes absolutely looks like a concrete monorail pillar. Same size, same shape, same color

problem is that with Tesla's super strong and high quality paint, people are probably justifiably scared their paint might leave permanent damage on the boxes.

Oh wait, typo. Opposite of everything above.

 

[Dan D.]

problem is that with Tesla's super strong and high quality paint, people are probably justifiably scared their paint might leave permanent damage on the boxes.

Oh wait, typo. Opposite of everything above.

Fine, your (/s) point is valid. Use $100 of square balloons then. Build some interesting obstacles. Have fun. I would. (Heck, if FSD Beta is allowed to be activated in Vancouver then come up here and I'll buy the balloons.)

22" White Foil Balloons | Squares | 50 pc

22″ Square White Foil Balloons by Anagram are great for promotions, decorations, events, and more! Create amazing Balloon Centerpieces, Balloon Bouquets, or use as Single Balloons attached to a decorative Balloon Weight that matches the color perfectly! Measures 22″ (diameter when fully...

www.balloonsandweights.com

There's no way you can convince me that the car 'knows it's not a real obstacle' just because it's made from thin lightweight material. It should not hit anything.

 

[Daniel in SD]

Sure. But DOES it see the pillars? It doesn't show them. All we know is that it seemed to veer towards them until the user manually swerved away. We don't actually know anything!

Going to have to wait for the button. I'm sure there's someone dumb enough to find out.
My wild guess is that the "Vidar" neural net will see the pillars in most lighting conditions.

 

[mikes_fsd]

Generally speaking, because they are being given access to things that people who paid before them are not,

How do you know when all these people paid for it?
And car majority are employees, that are helping to test this.

and that bothers me as it's just another way Tesla is screwing with customers.

Clearly you believe this is not ready or safe yet, but here you are "upset" that Tesla's customers are being "screwed" over because Tesla wants to make sure that the software is better before releasing it to more people!

So which one is it?

personally don't even care about auto steer on city streets, so it's not about me not having it. In fact it's very nearly worthless for me

Sure are invested in sh!tposting about something you claim not to care about!

 

[Ptheven]

How do you know when all these people paid for it?
And car majority are employees, that are helping to test this.



Clearly you believe this is not ready or safe yet, but here you are "upset" that Tesla's customers are being "screwed" over because Tesla wants to make sure that the software is better before releasing it to more people!

So which one is it?


Sure are invested in sh!tposting about something you claim not to care about!

I don't have much else to do

 

[GlockWorkOrnge]

I view it as doing a useless impressive thing much more cheaply. I just don't see much value in this sort of partial autonomy. It seems unlikely that they'll be able to get to real autonomy with the current hardware when so many other companies have invested even more money and engineering resources and achieved only extremely limited FSD.
Autonomy day was April 2019. Can we quantify the progress since then? How much further do they have to go?

When you say "partial autonomy," do you mean FSD or regular AP? I only have regular AP and it is incredibly useful on long or medium drives.

AP is only "partial autonomy," but I find it very useful.

 

[Daniel in SD]

When you say "partial autonomy," do you mean FSD or regular AP? I only have regular AP and it is incredibly useful on long or medium drives.

AP is only "partial autonomy," but I find it very useful.

I mean beta FSD. I also find TACC and Autosteer useful.

 

[emmz0r]

LOL. That is my daily commute!
@mikes_fsd - how to you justify this as a mind blowing experience when a full visual system can't even avoid massive concrete pillars? Edge case?

It wouldn't be mind blowing, but rather mind smashing if he didn't intervene...

Does the visualization even have something for columns/pillars? You can't really tell if it can't "see it" as it seems a lot of what the perception engine sees is not shown in the visualization. Sometimes it is the path planning that goes wrong even if the perception is correct (for example the fence example shown further up thread where it renders the road correctly, but the path planning still tells it to go offroad into the fence). In this case, it seems to detect the solid white lines correctly, but path planning still tells it to cross them.

Good question! It can certainly see edges, but clearly it doesn't see those or know how to react to it

 

[mikes_fsd]

Does the visualization even have something for columns/pillars? You can't really tell if it can't "see it" as it seems a lot of what the perception engine sees is not shown in the visualization. Sometimes it is the path planning that goes wrong even if the perception is correct

One of the big gripes from Chuck is that he misses the details that the dev view had (old FSD visualizations).

But just like the draw-bridge in his new video, pillars are not something that is rendered
Neither are street lights -- but stop signs and speed limit signs are rendered on their poles.

 

[gearchruncher]

But just like the draw-bridge in his new video, pillars are not something that is rendered

The issue is that they're not avoided either! The Seattle video shows the car trying to drive right into them more than once.

 

[GlockWorkOrnge]

I mean beta FSD. I also find TACC and Autosteer useful.

Yeah, AP and Autosteer as currently constituted are just great. If the FSD becomes that smooth and there is a subscription option, I will try it.

 

[stopcrazypp]

The issue is that they're not avoided either! The Seattle video shows the car trying to drive right into them more than once.

Yeah it does look like it is looking to go right into them, but then on the other hand it is legal to go between the pillars to get to the other lane (I believe it was trying to do a left turn). It's just a weird situation. But no one will be willing to risk trying that unless the visualization actually shows the columns, given there is no way to tell if the system sees it (or even if it sees it, will it make the right move to avoid them).

 

[gearchruncher]

People need to keep in mind that the human accident rate is about 1 per 150k miles

FYI, Tesla's own numbers put accidents at 1:484K miles, not 150k.

By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 484,000 miles.