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 don't know for a fact it would have hit the pillar.

Yes, I do know for a fact. In the video, we clearly see the car swerve toward the pillar and the driver goes "wow" and jerks the wheel back. So yes, it is clear in the video that the car would have hit the pillar if the driver had not intervened.
 
I direct quote someone calling people idiots, and I get blamed?
Your bias is showing. Show me where I ever called people idiots without quoting someone. "Idiots" is the term Tesla supporters use to dismiss anyone that gets hurt using AP so that they can ignore the mistakes it makes. It's not used by people that actually think real Human Factors Engineering includes considering real, broad, human behavior instead of relying on your system only being driven by expert drivers and beta disclaimers.
I'm not biased. I just know my safety is my own concern. You invoked your professional certification (which again, I don't care at all) at some point to imply that your word is the truth. So that, yeah, is basically calling others ignorant, or idiots.

When I took my Driver's Ed course (FWIW, I'm from Brazil and had a driver's license there, too), they always stated multiple times that your safety is your own concern. Especially on motorcycle training. You're the vulnerable one, so it doesn't always matter that you're right. You can be right AND dead, which is no good. In short, I have 4 Driver's Education courses in my past, and this is not bragging or anything, it's just a fact, and I'm sharing my own experience that in all of them, they make you understand that it's your responsibility, as the driver (or pilot), to keep yourself, your passengers and everyone around you safe.

Until the regulations change, this is the truth, not from me, but from the entities governing the roads. No assist / accessory is to be trusted more than yourself.

Yes, idiots exist. They always will. Cars go up to, I don't know, 200+mph today. Why? Should we revoke those, because there are idiots who will street race?

My answer is no. Let's give people what they want. Most are not idiots. And just like how fast your car can go, FSD is just another feature that requires common sense and responsibility.

Ultimately it's the Tesla haters that want to blame everything on Tesla, while ignoring equivalent issues that would (under the same train of thought) direct them against other manufacturers. I am not a Tesla fan. I am Tesla neutral. Just like I am with every other car manufacturer. Except Fiat. I don't like Fiat.
 
So yes, it is clear in the video that the car would have hit the pillar if the driver had not intervened.
Curious, would you say the same thing for a related situation where FSD beta doesn't seem to handle static objects well? The visualization seems to give a clear indication that it would hit the barrier.

path into static.jpg
 
Yes, I do know for a fact. In the video, we clearly see the car swerve toward the pillar and the driver goes "wow" and jerks the wheel back. So yes, it is clear in the video that the car would have hit the pillar if the driver had not intervened.

But those are not from V9 so you don't know what V9 is seeing.

EDIT: Just noticed this video had been previously posted by @diplomat33 (of course, haha). Anyway I'll just leave this here.


Here's a more informed discussion about this which clarifies what may be happening and the future prospects for this. It may be a labeling issue, etc. In any case production firmware (with the vision-based depth map) is able to identify the pillars as non-drivable space while the V9 beta (apparently) could not. It's not clear what elements of the vision-based depth map are being used in V9:



And in the replies, some discussion/speculation from @verygreen (there is more discussion than just this tweet - you have to scroll up and down for the replies):

 
Last edited:
i can’t rest, text, watch a video, or anything with AP right now, and I love it. It makes my long drives way less stressful because I can let the car do the tedious tasks while I monitor it and the road. I constantly disengage it and take over when it approaches situations I’m not comfortable with and it still makes my drive easier. You don’t need the system let you check out completely to be useful. It’s like cruise control, just letting the car handle maintaining speed on a highway was a big boon even though you still had to pay attention and drive because there is still an energy and mental drain from even the simple act of keeping the gas pedal pressed down to the same level that you don’t notice until you don’t have to do it anymore. I maintain situational awareness and keep my hands on the wheel and can feel by the torque of it whenever AP is going to do something weird (like aborting a lane change halfway through, for example) and can just easily override it, and it makes my drives easier. There is a real difference between moving the steering wheel yourself and letting the steering wheel move your hands, even though the actual motion is superficially the same.
I see you on that, and feel the same.

But city driving with all it's chaos and complexity is so different from highway driving. FSD beta is so far away from anything like AP on a uncompicated highway.

I wish they focused on getting to level 4 highway first instead of city driving.

The question between the lines were: what is proper use of FSD beta 9?
 
It also marks non-drivable space and the previous videos (see the first one you quoted) show it was able to mark the pillars as non-drivable space. It's unclear if this was not the case for the FSD v9 video (as we don't have the same raw output) or if simply the visualization not showing everything (which we know to be the case in general already, just not sure for this specific case).
Looks so, and is also my experience with AP2 on winding local roads and city streets back in 2017/18. The system tries to hard, and will try to drive where it really should disengage because capability is not there yet. From a safety perspective it would be better that it understood that these pillars -> anomaly -> disengage. Unprotected left -> no way, disengage.

It's like the noob in the snowboard park aiming for the biggest big jump because they saw the pros did it.They don't even have an idea of the consequence of a fail.
 
  • Like
Reactions: pilotSteve
Yes, I do know for a fact. In the video, we clearly see the car swerve toward the pillar and the driver goes "wow" and jerks the wheel back. So yes, it is clear in the video that the car would have hit the pillar if the driver had not intervened.
It could have also been attempting to go between the pillars. From the video it sure looked like was going for the pillar, but that remains a possibility (given apparently that is legal and from the videos linked above at least the previous version marked those areas as drivable areas). I personally would have been shocked by the attempt too given no visualization plus me not knowing it was legal (until reading the article linked previously), not sure of driver was aware either.
 
You are not answering the question between the lines: what is proper use of the FSD beta? Why do you want it?
My reasons are irrelevant.

Proper use is following the freaking warnings to be ready to take over.

Why do you continue to ask a very dumb question?

When it's open for wider beta, it will be improved and it will come with warnings and caveats that people with IQ higher than body temp (leave it to you to use F or C) must read and heed.
 
That can't be the way it's supposed to work. If that were true then any new object that appeared, or one that was positioned in a novel manner (e.g. a smoking, upside down crushed car), wouldn't be avoided.

that probably is how it works. There are a lot more things the car doesn’t know know about then there are things it knows about. I think if you set it up to avoid everything it doesn’t know you would have a lot more phantom breaking. It assumes (correctly) that the driver will intervene if the unknown is something to worry about.
 
  • Like
Reactions: ahoen117
I wish they focused on getting to level 4 highway first instead of city driving.
Sort of off-topic, but I'm not sure that is possible. There are rare transient situations in highway driving which are extremely complex (to some extent it depends on your definition of highway, but even freeway driving has exceedingly difficult situations which arise, rarely). I certainly understand your sentiment (and agree it would be great if we were seeing huge improvements there!), but I feel you have to solve the entire problem to even have the prospect of level 4 highway driving. The distribution of task difficulty is shifted downwards for highway driving for sure, but the outliers on the distribution are still there, and probably similar to the difficulty level of quite difficult but not that uncommon city driving.

Even more off-topic:
For level 4 highway driving, passing on a two-lane highway seems like a super difficult problem best left to humans (who also can't do it that well)! It's also the most stressful and one of the most dangerous parts of driving a two-lane highway, so it would be essential to get that working for level 4 highway driving (people aren't just going to sit behind someone going 10mph under the speed limit). I guess you could argue that it's not essential and you just wait for a passing lane but I think that's a pretty huge limitation and makes it more like L3 if takeover is required in those cases in order to accomplish the pass.

The videos have been interesting. I am interested in more highway/freeway videos of beta performance, because it seems like there are claims that it really is more smooth on the highway. I'm not convinced by these claims yet (I want to see more videos somehow comparing the behavior of the production and beta in freeway traffic) but if they exist, it would be pretty nice if some of these improvements were merged into the production release flow, even if we don't get The Button or The Beta. After the 2600-mile recent road trip, I can report the latest production release is still pretty awful, though much better than it has been in the past. It just has nearly no ability to reduce jerk. If you're reacting to someone who has stopped or is turning off on the road ahead, it's possible to slow down significantly with extremely low jerk. Tesla TACC consistently fails to do this, instead opting for a more maximum jerk strategy (they approximate a delta function of jerk as far as I can tell). (Yes, I know I am arguably using TACC outside of its ODD as described in the OM - I don't keep track of the exact verbiage there, nor do I care, since there seems to be tons of ambiguity last I checked - but that misses the point, since the same argument can be applied to the freeway.)
 
Last edited:
  • Like
Reactions: Jeff N and daktari
For level 4 highway driving, passing on a two-lane highway seems like a super difficult problem best left to humans (who also can't do it that well)! It's also the most stressful and one of the most dangerous parts of driving a two-lane highway, so it would be essential to get that working for level 4 highway driving (people aren't just going to sit behind someone going 10mph under the speed limit). I guess you could argue that it's not essential and you just wait for a passing lane but I think that's a pretty huge limitation and makes it more like L3 if takeover is required in those cases in order to accomplish the pass.
This is not me disagreeing with you, I think your point is 100% right. But I'd like to share my own experience. I've been behind that car, 10 under limit, kind of mentally cursing them. But I was on Autosteer. And honestly? Besides for a little while, I didn't care. I didn't care enough to override and pass nor did I care enough to actually mentally curse at them any longer. They were setting my speed limit and I was happy to not drive and just look at my car to do it's thing.

My point is, although you're right (that's a lame speed to drive), NOT actively driving switched up my expectations. It was a different scenario and FUN.
Therefore, NOT actively driving might, in the long run, make us care less about road time, Perhaps we only care about driving time after all. Food for thought!
 
  • Like
Reactions: GlmnAlyAirCar
I wish they focused on getting to level 4 highway first instead of city driving.


I am 10,000% in agreement with you. The overwhelming majority of the driving I do that is frustrating is on the highway. I could barely give less of a crap if I *ever* get true door to door operation as long as when I'm on the highway I can sleep or read or rub myself in peanut butter and stick cocktail stirrers up my nose if that's what I feel like doing.
 
Sort of off-topic, but I'm not sure that is possible. There are rare transient situations in highway driving which are extremely complex (to some extent it depends on your definition of highway, but even freeway driving has exceedingly difficult situations which arise, rarely). I certainly understand your sentiment (and agree it would be great if we were seeing huge improvements there!), but I feel you have to solve the entire problem to even have the prospect of level 4 highway driving. The distribution of task difficulty is shifted downwards for highway driving for sure, but the outliers on the distribution are still there, and probably similar to the difficulty level of quite difficult but not that uncommon city driving.

Even more off-topic:
For level 4 highway driving, passing on a two-lane highway seems like a super difficult problem best left to humans


But APs ODD is divided limited access highways where that's never a thing it has to do.

I think getting L4 in THAT ODD should be pretty easy.

1) Don't hit parked/stopped things partly in a lane/on shoulder.

2) Better handling of places they don't properly mark exit/merge lanes with dotted lines to understand the existing lane didn't suddenly double in size (around here they DO mark em properly, but I've seen this issue reported as a problem in certain states where they're not so good at it).

3) Be capable of safely pulling over to the shoulder if you run into something you can't handle.

4. Some handling of major road debris.

That'd pretty much cover it.


City driving is massively more complex because you have intersections, cross-traffic, oncoming traffic, parked and double parked vehicles are common, as are pedestrians and cyclists, and sometimes things like highly condition signs like where speed limit changes during school hours, or parking rules change by date/time, etc.
 
  • Like
Reactions: Microterf
But APs ODD is divided limited access highways where that's never a thing it has to do
Yeah I know (as I said!!!). I’ll say also that are plenty of places where AP is currently allowed to be used in full which are not limited access (though that could be changed of course). US-395 divided highway segments for example. Anyway, even with your listed limitations, and even narrowing to freeways only, I still think there are corner cases that would be harder than you might think. It is all about the distributions. You can’t crash. Driving is incredibly complex - a lot more complex in the city no doubt - but the very rare outliers are important!

Think through all the craziest things you’ve seen when driving.

I’m just not convinced you can get L4 highway and have that be an appreciably easier problem than solving city driving. It’s easier of course, just like it is for humans. But for humans they are actually remarkably similar tasks - just a bit more annoying complexity, as you listed - but the fundamentals are very similar - don’t hit anything and follow the rules! The idea that if Tesla just focused on L4 highway they could be there quite rapidly, much more rapidly than city driving - not convinced the timelines would work that way. The big problem as I see it is the perception - and beyond that, really understanding what is happening. That seems really difficult, no matter what the domain.
 
What....have you watched this? What a joke... A simple unprotected left turn and it fails consistently and tries to cause a dozen accident and head on collisons.
You do realize that performance includes safety right?
Here are 122 unprotected left from one rider (@JJRicks wondering when you are going to post the compilation)

versus v9



If this is the performance of your so-called leader of AV that is supposedly 5-10 years ahead. Then we will never solve AV.
The first sentence on that video description says this:

"FSD Beta is really quite impressive. I've been driving on it since December 2020 and for almost 20,000 miles. This is generally a more negative video and I can see how it could be taken the wrong way. But Tesla and the Full-Self Driving technology is really cool, this video explores only a small area in which it needs improvement".

Yes left turns with fast approaching traffic at high speed is going to be the most difficult problem to solve (the hardest I would think). Tesla will get there. There has been a huge improvement in the last 5 months.
 
  • Like
Reactions: rxlawdude
The first sentence on that video description says this:

"FSD Beta is really quite impressive. I've been driving on it since December 2020 and for almost 20,000 miles. This is generally a more negative video and I can see how it could be taken the wrong way. But Tesla and the Full-Self Driving technology is really cool, this video explores only a small area in which it needs improvement".

Yes left turns with fast approaching traffic at high speed is going to be the most difficult problem to solve (the hardest I would think). Tesla will get there. There has been a huge improvement in the last 5 months.

I feel like he is downplaying the areas where it needs improvement. It is not just "one small area" that needs improvement. For one, doing left turns with fast approaching traffic is not a "small area", it is a pretty major driving task that humans have to do all the time when driving in cities. Also, FSD Beta has more problems than just not being able to reliably do left turns with fast approaching traffic at high speed.

Exhibit A:


0:13 First Mistake
2:34 Disengagement
3:12 Car Hits Bush
3:38 Wrong Lane!
4:50 Left Turn Fail
5:25 Near Collision
9:12 Disengagement
9:25 Disengagement

Exhibit B:

Car almost hits large pillar at 17:06.