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.
Tesla has finally achieved the FSD promo video from 2019! Mission accomplished!

Similar kind of drive:


Watching this, I'm further convinced that if FSD beta can handle New England (or Northeastern US), it can handle anything nationwide. The roads in this video are too tame.

But good to see FSD doing well in certain areas. Based on my travels, most of the country have reasonable roads and intersections. NE is just a big clusterf*ck.
 
  • Like
Reactions: aronth5
Watching this, I'm further convinced that if FSD beta can handle New England (or Northeastern US), it can handle anything nationwide. The roads in this video are too tame.

But good to see FSD doing well in certain areas. Based on my travels, most of the country have reasonable roads and intersections. NE is just a big clusterf*ck.
Some parts of SF can be quite difficult - but I’ve not been to NE. Are there any NE FSD youtubers?
 
Wholemars has a long raw video. He kept driving until he had to take over. FSD drove for 70 minutes without needing to disengage … So, I guess, Mobileye/Cruise and others showing hour long videos isn’t a big deal after all :)

Correct, an hour of driving is meaningless. What is the maximum YouTube video length?
Waymo should publish their 60k mile driverless video. I guess that would be about 3,000 hours? They were rear ended once in that time. It would be interesting to see what precipitated that collision.
 
Tesla has finally achieved the FSD promo video from 2019! Mission accomplished!

Similar kind of drive:

I couldn't do this drive without taking over at around the 1:30 mark due to the abnormally long time to take a right on red unless it was a no right on red. There was a car behind them so I'd be really self conscious.

I also would have taken over at 10:45 which the driver essentially did, but that was construction so its outside of expectations.

In my quick testing of 10.9 I had the following while testing it at night without anyone around.

1.) It took all the corners from stop lights/signs without me taking over. Now it didn't do any of the lefts smoothly as it seems to do a sudden twitch at the start then a couple correction changes so its never a smooth lane change. It did manage a couple smooth enough right turns. In a couple cases it still got too close to barriers. I reported them, but didn't take over.

2.) I didn't have much in the way of phantom braking or sudden twerks on the steering wheel while going straight. I could see not holding the steering wheel, but I'm too weary of its previous steering twitches to do that. There were some small nuisance momentary pauses where it decreased speed a bit and corrected. Mostly in slight corners.

3.) It did an auto-lane change in the middle of an intersection which is a bit bothersome.

4.) It re-centered itself in the middle of the two lanes of a two lane each way highway (undivided) because the road is missing lane dividers (dots that got plowed away after a snow storm a few weeks ago). I tried to do a lane change while in the middle just to see what it would do. It headed for the middle turn lane before I stopped it. Rather amusing, and I didn't expect it to handle this case. I did send an email to the local country who forwarded it onto the right people to get the road fixed.

5.) drives too fast in residential areas. I wish there was a way for it to do 5 under the speed limit there, and the speed limit elsewhere. Having to adjust it quickly after it sees the signs is annoying.

10.9 is better than 10.8.1, but for me its still a "The driver is mostly there to hold the wheel and scream in terror" kind of experience.
 
Last edited:
The state of Tesla FSD

I made it to hyper change and then stopped it.

I'm not a strong supporter of the need for disclaimers about content. But, any hyperchange video should come with a "this video might induce vomiting".

I didn't watch the video so I don't know your motivation for posting it.

All I know is I can't stand the guy, and I'm not going to take a chance on watching it.

In my testing FSD has improved. But, there are so many fundamental flaws that I'm not sure it makes a difference.

Tesla doesn't have any feedback loop for getting issues fixed. As a result maps/navigation are often wrong resulting in a lot of manual changes.
The speed it selects for a road is done entirely wrong. It should be a max speed where the environment its in will dictate going slower, and it does so this some of the time. But, most of the time its like "lets get to set speed in this narrow space", and I'm like "No, lets not".

What I find absolutely ridiculous is even when it does a good job, and you're arriving at the destination it doesn't actually stop. It puts itself in TACC mode. Like WTF?

I like the measurement of spousal acceptance, and I don't think FSD Beta 10.9 is there yet. In fact maybe the best way of measuring FSD beta improvements is simply using an accelerometer, and setting maximum comfort level. Maybe that along with other analytical measurements would remove emotional elements from it.

I have an ADXL355 so I might record a timestamped drive, and see if there are any obvious failed trigger points.
 
I made it to hyper change and then stopped it.

I'm not a strong supporter of the need for disclaimers about content. But, any hyperchange video should come with a "this video might induce vomiting".

I didn't watch the video so I don't know your motivation for posting it.

All I know is I can't stand the guy, and I'm not going to take a chance on watching it.

In my testing FSD has improved. But, there are so many fundamental flaws that I'm not sure it makes a difference.

Tesla doesn't have any feedback loop for getting issues fixed. As a result maps/navigation are often wrong resulting in a lot of manual changes.
The speed it selects for a road is done entirely wrong. It should be a max speed where the environment its in will dictate going slower, and it does so this some of the time. But, most of the time its like "lets get to set speed in this narrow space", and I'm like "No, lets not".

What I find absolutely ridiculous is even when it does a good job, and you're arriving at the destination it doesn't actually stop. It puts itself in TACC mode. Like WTF?

I like the measurement of spousal acceptance, and I don't think FSD Beta 10.9 is there yet. In fact maybe the best way of measuring FSD beta improvements is simply using an accelerometer, and setting maximum comfort level. Maybe that along with other analytical measurements would remove emotional elements from it.

I have an ADXL355 so I might record a timestamped drive, and see if there are any obvious failed trigger points.
Since 10.8, I think I’m there with the spousal acceptance lol. I can drive many miles before she realizes FSD is even on. It’s generally smooth enough that it doesn’t bother anyone, but then of course at some point it’ll do something incredibly dumb like swerve immediately into another lane without reason, or jerk through a left/right turn.

Maybe it’s a case of overfitting in the SF Bay Area, but I can do most 10-20 mile drives in my area with 0-2 disengagements. In the early days of FSD 10.2-10.4, it was closer to like 1 disengagement every 2-3 miles.
 
The state of Tesla FSD

Lots of issues in this video. Car in left-turn lane at a red light while attempting to drive straight with right blinker on in traffic. Flipping between turning and not turning multiple times

He fast-forwards through a disengagement where it veered towards construction
Five.jpg
Six.jpg


Car changes its mind about blue path in mid-intersection
One.jpg
Two.jpg


Three.jpg
Four.jpg
 
  • Helpful
Reactions: daktari

Attachments

  • 1642625437643.png
    1642625437643.png
    1.6 MB · Views: 40
  • 1642625437880.png
    1642625437880.png
    1.6 MB · Views: 32
  • Ten.jpg
    Ten.jpg
    27.5 KB · Views: 21
  • Eleven.jpg
    Eleven.jpg
    11.8 KB · Views: 28
Last edited:
In fact maybe the best way of measuring FSD beta improvements is simply using an accelerometer, and setting maximum comfort level. Maybe that along with other analytical measurements would remove emotional elements from it.

I have an ADXL355 so I might record a timestamped drive, and see if there are any obvious failed trigger points.
This is a great idea. Compare it to your manual relaxed drive and manual agro drive and maybe wife's drive?
 
Since 10.8, I think I’m there with the spousal acceptance lol. I can drive many miles before she realizes FSD is even on. It’s generally smooth enough that it doesn’t bother anyone, but then of course at some point it’ll do something incredibly dumb like swerve immediately into another lane without reason, or jerk through a left/right turn.

Maybe it’s a case of overfitting in the SF Bay Area, but I can do most 10-20 mile drives in my area with 0-2 disengagements. In the early days of FSD 10.2-10.4, it was closer to like 1 disengagement every 2-3 miles.
I need to stop watching FSD Beta videos from Cali because its almost a night and day difference from what I experience.

Even if I was blinded and put in the passenger seat I could tell the difference almost immediately where I drive. But, in Cali it seems like it would take me a lot longer.

It's certainly better than 10.3.1 (the earliest I had), and in fact today is the first time in awhile I used on a left turn from a light with a vehicle behind me. It did fail which it didn't last night. I'm not sure why it briefly paused, but I can't be paused with someone behind me needing to get through the light.

If I was to compare my results with other regions it would mostly be the smoothness factor of turns at intersections. My want is a nice gentle turn, but what I get is 3-4 changes through the corner.
 
  • Informative
Reactions: momo3605
This is a great idea. Compare it to your manual relaxed drive and manual agro drive and maybe wife's drive?

For good or bad my personality "quirks" have kept me single so no wife data.

I am curious to see how the data compares especially when trying things. Like today I tried putting chill mode on before the drive to see if it would cut down a bit on the acceleration at times.

Some say it has no impact on TACC/AP/FSD/etc, but a couple months ago it seemed to rein in the over acceleration issue I was having with TACC during stop and go traffic while on AP on the freeway (the city streets with FSD Beta were okay, but something was up with AP on the freeway). With later versions of the FSD Beta build I didn't need to do that. I think it was some glitch with the Tesla Vision stuff.

It's an annoying setting to use because when I drive manually I want to go, and chill mode is like "umm, I'd rather not".

My expectation is my data will have less choppiness, and will have nice ramp ups.
 
n fact maybe the best way of measuring FSD beta improvements is simply using an accelerometer, and setting maximum comfort level. Maybe that along with other analytical measurements would remove emotional elements from it.
I'be been playing around with the "Sensor Play" app on iOS, which seems to be pretty decent and offers decent recording features (a few apps I tried were pretty bad). I would have liked a mic input also so I could correlate events to sensor data.
 
  • Like
Reactions: S4WRXTTCS
FSD Beta goes through a gap between two stopped cars and completes an unprotected left turn:


Good for FSD Beta that it pulled it off. But I don't think I would ever try that on FSD Beta. It just seems too risky to me.
The problem with this kind of maneuver is that there is no way to tell if FSD really handled it amazingly, or if it just got lucky. I frequently encounter events where FSD goes through an intersection and I am not convinced it can actually see where it needs to look. (Of course I know where to look and wouldn’t let it go if unsafe.) The same locations where it does “great” are often the same locations where I have to take over because it attempts to go when unsafe.
 
  • Helpful
Reactions: diplomat33