FSD Beta Videos (and questions for FSD Beta drivers)

[JulienW]

Where is the prediction? What specifically is the indicator? Curbs are always red and there is a cone here so it is definitely visualizing that.

Simple, the projected path of the car. If it follows the Blue prediction it hits the curb, which it did.

EDIT: Just to add as the curb went "out of view " of the front cameras it suddenly regresses farther back, probably meaning the Tesla misjudged it at the last second. Time stamp 11 second to 12 seconds.

 

[FLMan]

You can clearly see the car drifting away from the left line and out of the center of the lane when he starts saying "woah woah woah". I'm somewhat new to FSDb and I would never let my car drift out of the lane that far, even on a highway. It constantly wants to do dumb things and I have to shut it off.

The car is drifting because there's no line on the right. Even NOA does this on highways. It apparently doesn't know what to do about the curb coming into the road even though it "sees" it and marks it red.

 

[Matias]

Found it. Looks like it gets a lot of love.

Google Maps

Find local businesses, view maps and get driving directions in Google Maps.

goo.gl

View attachment 833733

Lol. As you said, FSD Tesla was not the first car to hit it..

Maybe the local wheel dealer has installed that!

 

[AlanSubie4Life]

To mee the visualization shows that it is going to hit that curb.

Looks like the car is "predicting" a hit

Yeah, I don’t really see it. Would need a bigger video.

 

[AlanSubie4Life]

Lol. As you said, FSD Tesla was not the first car to hit it..

Maybe the local wheel dealer has installed that!

View attachment 833734

Yeah it is completely terrible. But anyway the car should avoid it. It did visualize it, at least partially, but definitely not surprised it hit it.

Probably most of that love is from the busses turning into the bus stop; in general I would not expect through traffic to hit it though it probably does of course.

Will be fixed up in 10.13 I am sure. Only a few months left until completion.

 

[AlanSubie4Life]

drifting away from the left line and out of the center of the lane

Might have to do with the standard asinine behavior where there is still some bias to put itself in the middle of the available lane (in this case the bus stop ahead it thinks might be a lane so tries to use some of it). Maybe.

 

[qdeathstar]

Well, I think the car thought that you thought the car was going to avoid to curb so the car thought it was a good idea to visualize it hitting the curb so you wouldn’t think that thought. But you didn’t think about what the car was thinking and now your rim thinks you both are jerks.

 

[Matias]

Well, I think the car thought that you thought the car was going to avoid to curb so the car thought it was a good idea to visualize it hitting the curb so you wouldn’t think that thought. But you didn’t think about what the car was thinking and now your rim thinks you both are jerks.

 

[Daniel in SD]

I think the entire vision system rendering is biased towards showing low confidence perception and the planning system is biased towards high confidence perception. That way whenever it makes a mistake you can go back and watch the video and say "perception is great, all they need to do is fix basic path planning!" Nobody ever goes back and watches these videos to find all the phantom curbs you would have been swerving all over the road to avoid if planning responded to low confidence perception.

 

[Electroman]

That Ogan guy is a scum. Anyone can cherry pick and show only bad things, on a product still under development.

I remember TSLAQ highlighting and spotlighting only those clips where Autopilot had trouble in the initial days of AP2 development. And now AP is matured and is largely a finished product and is doing great.

Remember the whole autonomous driving is new territory and is not a 100% solved problem for anyone in the industry. There will be gradual progress, and these scaremongering scumbags like Ogan want to highlight only the shortcomings and create fear.

 

[Cheburashka]

Nah it's valid criticism. Elon has promised robotaxis etc for a while. The criticism is well deserved.

 

[Tam]

...I didn’t see anything in the visualizations indicating it was going to hit the...

Same principle as when the steering wheel went crazy by turning so fast to the right (blurring the steering wheel picture), aiming right at the parking truck on the right.

But the blue predicting path was not aimed at the truck.

 

[Electroman]

Nah it's valid criticism. Elon has promised robotaxis etc for a while. The criticism is well deserved.

Not at all valid. Criticize Elon's timeline all you will. But taking a dig on the yet to be solved edge cases and ignoring the incredible progress made so far, and using that to project the endeavor is a massive failure, that too for a problem no one else has solved yet, is cheap shot. To be expected from characters like Ogan.

If the moon project did not finish before the end of the decade as Kennedy proclaimed, but it took another 5 years, does that mean the project is a failure.

 

[Tam]

...product still under development...

During the 2022 Q2 Earnings Call on 7/20/2022:

"But we’ve got a team of about 120 people in our software AI group that are extremely talented. And I think we will have. I’m highly confident we will solve full self-driving and it still seems to be this year. I know people are like says that. But it does seem to be epic. It does seem as though we are converging on solving full self-driving this year."

This year means 5 more months left to clean up these kinds of mistakes.

 

[Tam]

...If the moon project did not finish before the end of the decade as Kennedy proclaimed, but it took another 5 years, does that mean the project is a failure...

The difference is after Appolo 1 fatal tragic fire accident, NASA wouldn't say the moonshot is this year, so those who want to watch the launch to the moon, pay $12,000 now or the price would be increased.

 

[SO16]

Elon should definitely stop with his FSD predictions. His history on this subject is poor.

That said, Ogan does have a history of only reporting one side to show negative bias on practically every Tesla topic.

 

[Electroman]

.. again.. whine and complain all you want on these unreasonable silly timelines from Musk.

But pointing out the still to be ironed out rough edges to project a false sense complete failure while completely ignoring the progress, reflects poorly on the person than on the program. All it matters now is, are we making progress? The answer undoubtedly is YES.

It can already drive without human intervention in a huge set of locations. To get that to 100% will happen sometime in the future. One year? five years? no one knows.

 

[Tam]

...One year? five years? no one knows...

It is absolutely fine to dream big and talk big.

GM 1939 Futurma already talked about self driving cars or 83 years ago.

GM's talks have been free.

Tesla's talk is different. You need to be on pins and needles because last Wednesday, it said the FSD "is basically currently ridiculously cheap, assuming FSD materializes, which it will." and if we ignore that, "Yes. We will increase the price of FSD sometime later this year."


Tesla talk costs money. Pay $12,000 or it will increase!

The first documented fatal Tesla accident on its automation system is 2016. On Sunday,7/24/22, there was another fatal accident that killed a motorcyclist:

Yet another, AP fatality under investigation

https://www.inputmag.com/tech/tesla-autopilot-crash-federal-investigation-nhtsa/amp

teslamotorsclub.com

Tesla collision avoidance technology deficiency is the same resulting in fatalities from the days of AP1 to now.

 

[Electroman]

Imagine the Ogans of the early 1900s would be calling to cancel air flights because of the accidents that happened during those initial days.

 

[momo3605]

But pointing out the still to be ironed out rough edges to project a false sense complete failure while completely ignoring the progress, reflects poorly on the person than on the program.

Honestly I believe that video is valuable. If I had never used FSD beta and only watched wholemars videos, I would be lulled into a false sense of security and likely have a far more fatal accident.

Thanks to videos like this, I know to be careful when using FSD in certain instances and disengage when it’s getting close to curbs.