OFFICIAL BUTTON WATCH

[tomas]

Pontificating is generous.

 

[powertoold]

I'm at 100% Tesla will achieve avg human level safety with their fsd beta approach. I'm not sure what others are seeing. Many here would like to see Tesla fail at fsd so they can say I told ya...not enough sensors, needs lidar, Elon lies, etc.

To me, there's no more question about whether Tesla will achieve it, and also I'm sure it'll be within a short period of time, shorter than any of the other developers can scale by a couple years.

 

[Backup Account]

I'm seeing what you're seeing, except I'm also a software developer and can see potential. What the fsd beta is showing us is that given enough data/training, vision alone is adequate to identify and maneuver essentially all driving situations. If there are visual cues needed to perform a driving task, then Tesla will eventually be able to deal with it. This is what Elon means when he says they are no more remaining roadblocks for FSD.

I would only be concerned about FSD beta failures that aren't related to visual cues. So far, I haven't seen these examples. If you know of any, show us a video.

The problem with your comment and a lot of other doubters on this forum is that you're always looking at the current failures. What I'm looking at is whether the approach in general will eventually lead to success. There are certain aspects of fsd beta (like the left turns I mentioned) that demonstrate a likely success for a vision-based approach.

Exactly. I'm focused on looking for where the perception system is failing (vs. planning / mapping etc...) because that is the part that is perceived as "limited" by only having cameras.

If the perception system is working great, there's no reason Tesla doesn't have the potential to improve the other aspects to be better than competitors.

 

[qdeathstar]

what weirds me out is how anyone takes anything he says about FSD seriously anymore. The cognitive dissonance must be amazing.

the correct point is that no one has level two released right now and they all say it is right around the corner....

 

[qdeathstar]

I'm at 100% Tesla will achieve avg human level safety with their fsd beta approach. I'm not sure what others are seeing. Many here would like to see Tesla fail at fsd so they can say I told ya...not enough sensors, needs lidar, Elon lies, etc.

To me, there's no more question about whether Tesla will achieve it, and also I'm sure it'll be within a short period of time, shorter than any of the other developers can scale by a couple years.

Which facts?

 

[Cheburashka]

I would only be concerned about FSD beta failures that aren't related to visual cues. So far, I haven't seen these examples. If you know of any, show us a video.

I hope you are right. To you "visual cues" note, I think it's much more than that. Part of what I'm seeing is poor path planning and prediction due to lack of data.

Take this example at 11:59. The car nearly gets T-boned because it doesn't see the cross-traffic adequately and does not react in any way to it. It only sees one car (SUV), but ignores the one right after.

 

[powertoold]

I hope you are right. To you "visual cues" note, I think it's much more than that. Part of what I'm seeing is poor path planning and prediction due to lack of data.

Take this example at 11:59. The car nearly gets T-boned because it doesn't see the cross-traffic adequately and does not react in any way to it. It only sees one car (SUV), but ignores the one right after.

FSD beta 8.2 doesn't drive well in every situation right now, but it's not because the sensor suite or approach is inadequate (especially in the video you posted). If you take the first beta release until 8.2, you can see the rapid pace of improvement and potential.

 

[Cheburashka]

FSD beta 8.2 doesn't drive well in every situation right now, but it's not because the sensor suite or approach is inadequate (especially in the video you posted). If you take the first beta release until 8.2, you can see the rapid pace of improvement and potential.

I don't know about that. I think it's pretty good in suburban environments as is, but it's still pretty terrible in many urban situations. I have no idea why Tesla can't just geofence these places out and give us something we can use.

In the example I provided, it's pretty clear that the sensor suite is not enough for this kind of situation. It doesn't appear that the car sees the cross traffic and nearly gets t-boned.

 

[powertoold]

In the example I provided, it's pretty clear that the sensor suite is not enough for this kind of situation. It doesn't appear that the car sees the cross traffic and nearly gets t-boned.

I see blue boxes on the right in the visualization, for both the first and second car.

If anything, it would have been better had the car waited 2-3 seconds before proceeding through that intersection. Fsd beta has historically been bad at two-way unprotected lefts (your video is a similar situation, although the car isn't making a left, but going straight through a two-way intersection without a light is also more risky for humans).

 

[Cheburashka]

I see blue boxes on the right in the visualization, for both the first and second car.

If anything, it would have been better had the car waited 2-3 seconds before proceeding through that intersection. Fsd beta has historically been bad at two-way unprotected lefts (your video is a similar situation, although the car isn't making a left, but going straight through a two-way intersection without a light is also more risky for humans).

The space between the Tesla and the "second car" seems to be pretty wide compared to what was captured by the repeater cam. If that is indeed the car that almost T-bones the Tesla in that blue box, then there seems to be an issue with perception.

 

[CDN-Build]

I'm at 100% Tesla will achieve avg human level safety with their fsd beta approach. I'm not sure what others are seeing. Many here would like to see Tesla fail at fsd so they can say I told ya...not enough sensors, needs lidar, Elon lies, etc.

To me, there's no more question about whether Tesla will achieve it, and also I'm sure it'll be within a short period of time, shorter than any of the other developers can scale by a couple years.

Which facts?

I don’t think his post requires facts ?
He stating his opinion.

 

[qdeathstar]

I don’t think his post requires facts ?
He stating his opinion.

the earth is square.

 

[eggzrule]

the earth is square.

You forgot to provide your evidence The International Square Earth Society.

 

[Silicon Desert]

View attachment 653845

hahaha! We have a new timeline here folks, and it's even worse than today's earlier timeline of "hopefully", we're now ASPIRATIONAL

Hmmmm, I better go get my dictionary..... "Aspirational". He might have a different view of the meaning

 

[Silicon Desert]

If I have thrown heads 9 times of the last 10 throws that's a result of 90% heads. 90% is not the probability of heads, it's still 50%.

I get your point Dan, yet not a good example. You know you will be flipping for a LONG LONG time to get that outcome ...... UNLESS maybe you have heads on both sides of the coin and the 10th time it rolls off the table.

 

[Webeevdrivers]

According to Elon Tesla will have theirs by June.

Funny. . And yet sad at the same time.

 

[qdeathstar]

Funny. . And yet sad at the same time.

Let me know when someone releases their version, not limited to a small number of roads, then we will talk. Till then if elon squeezes his out in june or july or september he will be the first.

 

[Silicon Desert]

THE BUTTON PREVIEW...
We actually get 3 buttons. Select your level of autopilot risk

 

[Cheburashka]

I knew it was coming with 2021.4.20!

 

[qdeathstar]

THE BUTTON PREVIEW...
We actually get 3 buttons. Select your level of autopilot risk

View attachment 654435

Definitely gonna hit up the question mark.