Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Poll: your gut expectations for FSD v9

I feel like the improvement from FSD v8.2 to FSD v9 will be…


  • Total voters
    119
  • Poll closed .
This site may earn commission on affiliate links.
During a year of driving, I'll probably make a "Chuck" left (that kind of risky unprotected left) 5-7 times.

Ok? I have two each way on my way to work every day without lights at all, and two more with traffic lights. Somehow I've done them way more than 70 times in my life without an issue.

But this is irrelevant. You can't say something is L3+ as long as it isn't used very often. This isn't the lumbar system in a seat.

Unprotected left turns are a needed function to navigate many places. Your system either does them reliably enough to be at an L3 level, or you can't use them. Tesla is not going to be able to say "we have L3+, as long as we can navigate you there without a unprotected left!". People already knock competitors for preferring right turns. And it's not some sort of credit system- Unprotected lefts cause a collision every 1:10000 events, so you're only allowed to have your Tesla do it 7 times a year. L3+ will be rated at a per mile or per hour basis across the whole fleet- and it will not surprise me to hear the average car does an unprotected left every 10-100 miles (lots of entrances to residential areas look like Chuck's videos). As just one possible failure in those 10-100 miles, it needs to be damn reliable if you want 1:1B mile failure rates, which is where I got my 1:100M-1:1B guestimate from.

This 70%+ issue is a HUGE issue for L2 as well. Let's say it gets to 99% like AP is today. Yeah, you're supposed to pay attention, but let me just look at my phone for a second, it's worked every time before, when it decides to make an unprotected left. 1:100 chance of major accident isn't too bad, right?
Elon claims current AP is already better than humans (99.99999%), yet we have a few people die a year on AP as they are lulled into complacency. Can you imagine how bad it would be if AP was 99%?
 
Last edited:
During a year of driving, I'll probably make a "Chuck" left (that kind of risky unprotected left) 5-7 times.

If it can't do unprotected left turns across multiple lanes, it's useless for me. Avoiding them would add a completely unreasonable amount of time and distance to my commute in both directions.
 
I think 6 lane unprotected lefts with 50mph+ cross traffic are rare. Tesla can just avoid them.
6? Chuck's videos are crossing 3 lanes. The speed limit on that road is 45MPH per the display.

And your argument is that Tesla will do 70% for 3 lanes, but will be 99.999999% fine at 2 lanes? Or fine at 35 MPH but not 45 (as long as nobody is speeding of course).... As a human, do you really find a 3 lane crossing 10,000X as hard as a 2 lane?

You cannot put all these caveats in and then suggest that Tesla has really cracked FSD and is leading in this space. Superhuman is superhuman.

Tell me, if Tesla can just avoid them, how far would Chuck's car have to go out of the way to avoid an unprotected left? In many states, left turns at intersections can be unprotected. Do you really think a car that just sits there, blocking traffic, waiting for a green left arrow will be accepted?

You wanna see one of my unprotected lefts?
Look all up and down that road- it's covered in unprotected turns.
 
You wanna see one of my unprotected lefts?

Ya, those look a lot easier than Chuck’s.

Chuck’s roads may be 45mph, but that means cars are going 50-55mph.

also, 6 lane road is 3 per side:

41583A9A-EF05-4404-975C-9FBC9B449363.jpeg
 
Ya, those look a lot easier than Chuck’s.
So, if Chuck's hits 70%, what should those roads hit if they're so much easier? Will it be good enough that I won't have to grab the wheel once a week?

I would have thought a good vision NN would just be able to see each car, estimate time to arrival, and calculate how long it would take to cross. Chuck's videos are often going to collide with the first, nearest lane. Why does a NN that has to calculate EVERY CAR and EVERY LANE all the time care how many cars and lanes there are? It's superhuman. It pays attention all the time and never gets distracted.

Since you're so sure that this is rare, here's the turn into a Tesla "dealership" closest to me. Just a normal suburban road in Seattle. 3 lanes each way, turn lane, lots of businesses with entrances, and cross streets with no lights. (Old google photo before it was Tesla). And yeah. 45 MPH speed limit. Major arterial that connects Seattle to areas north. Already had to go here 5 times this year alone for service, so my quota is used up apparently.

FYI, average traffic on that road is 35,000 cars a day each direction, so a car every 2 seconds on average.


And here's another Tesla store near me, with a very short sight distance:

Again, only argument here is that 70% on Chuck's videos does not mean we're anywhere near L3 FSD on city streets, and we're not near 70% on Chuck's videos, and Chuck's videos are not extreme edge cases.
 
Last edited:
  • Like
Reactions: diplomat33
So, if Chuck's hits 70%, what should those roads hit if they're so much easier? Will it be good enough that I won't have to grab the wheel once a week?

Chuck’s turns are a lot more difficult because they also have to deal with cross-traffic cars turning left, an island, and trees / bushes obstructing oncoming cars. (Pic below from his video)

43E8AC5D-65B7-4334-AC75-4A1DD74EC001.jpeg
 
Chuck’s turns are a lot more difficult because they also have to deal with cross-traffic cars turning left, an island, and trees / bushes obstructing oncoming cars. (Pic below from his video)
If humans can do it with vision, you can see the cars. Static, non-blocking things don't matter at all.
This is the image from his car and from above. Nothing is blocking the cameras. It's a very clear, open road with good sight lines.

1622588377327.png
 
Last edited:
Ya, hopefully V9 is released soon, so we can see Chuck testing it. Until then, we'll just be spending a lot of effort discussing hypotheticals.
It seems to always degrade to that. "I'll be thrilled when it's 70%" as if there is any data to support that, or as if it means Tesla is really moving forward in FSD, or as if it is a critical metric. But then when any pressure is applied to that, it's "well I only have to turn like that 5-7 times a year", and then it's "well, this is a worst case", and then it's "well, this is all hypothetical, let's just wait."

...And then when the video comes out, and there's some not so great stuff in it? It's an early release.... Let's just wait some more.... The next one will be mind blowing... Elon's getting better, I think.

All most of us skeptics are saying is yes, there will be progress. But it will and has been VERY slow. There is no one sudden change that is going to get us there, blow our minds, or suddenly prove that Tesla has leapt ahead of everyone else. Moving to vision only is going to cause as many regressions as forward progress on the first release, and we don't have that first release yet. Any useful L3+ system is years and years away.

So maybe Tesla shouldn't be taking money for it and saying all their cars have all the hardware they need, and hyping themselves as the unquestionable leaders in this space, when we'd all be excited if they would only kill 30% of the people that use AP to cross a road, instead of 70%. Progress, yes. Mind blowing progress? Meh. Looks a lot like the normal development process of a very large, difficult program, not a company suddenly accelerating up the magic, mythical neural net S-curve well ahead of everyone else.

The one reality the skeptics operate under? It's that we've waited for the release in the past. And it hasn't been mind blowing, or fire like Elon says. It's a good, nice incremental increase that doesn't match the hype, and sometimes sends things backwards. So guess what? The next one probably isn't going to either. History is data, and we're just following that trendline, the one that actually exists instead of the one we keep getting told is right around the corner. From that point of view, people that say "just wait for the next one, it will blow your mind" look a bit odd.
 
Last edited:
Welp. Pack this thread up everyone. Elon says it's already superhuman. Obviously Chuck can't make these turns himself, or he's not a human. 😀

My baseline is pretty simple. If it appears in the US driving manual for my state, the vehicle needs to be able to successfully perform the operation or it's not even close. If the car attempts to jerk out into oncoming traffic like the latest round of 8.3 we say months ago now, then it's an instant fail. The number of lanes argument is absurd on its face, because either the car can identify oncoming traffic, or it fails. One lane, three lanes, ten lanes. It doesn't matter. This is extremely basic stuff, and attempting to split these hairs just shows how desperate people are to make the claims from Elon make any sense at all.

I've been waiting two weeks for several months now. I'm ready to be proven wrong any time Elon wants to push that update.