As others have reported since V11 FSD is much better in heavy rain including driving at night. FSD seems to see much better than I can. Whether that continues is another question. Before V11 FSD would simply disable itself in most rain conditions.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: FSD Beta tweets
- Thread starter Terminator857
- Start date
I agree that it's a lot more resilient. It may "work" (better) when degraded to some extent, but it really doesn't matter if the goal is autonomy. There is several order of magnitude missing in terms of reliability even in optimal conditions. An L2 may to disable at say 90-95% reliability whereas an L3 will need to hand-over at perhaps 99.9999%.
No, it works...it just reduces your max speed based on conditions.
FSDB on v11 works fine in heavy rain or the dark, even with the degraded message, which is more of a warning than anything else.
It drives when even Florida drivers pull over due to torrential rain. It's quite impressive.
I haven't seen 11.3 or 11.4 disable due to any conditions.
That's not what you said. You put "work" in quotes. It works fine and your comment about cameras not working at night or in rain without cleaning is not accurate.
Stop cherry picking parts of my sentences? It _works_ for L2.
This was the first point I made:
Cameras arent deemed safe enough even in optimal conditions. At night and in rain without cleaning?
This isn't factual. The current Tesla system has advanced, with only cameras, to perform about the same in adverse conditions as without.
There's no current evidence that night or adverse conditions are an absolute limiting factor in development. Currently it's software and potentially camera placement, but that's a poor argument that because no one has done it yet, that it can never happen. By that standard no one will ever reach level 5.
Of course it's factual. It's 100% correct. There are no L3-L4 vehicles in the world that uses vision-only. It's also true that a system will be degraded at low-light conditions and in rain.
It seems to be very hard for people to grasp that a system that needs to drive without a human OEDR safety net needs to be several orders of magnitudes safer than a system that doesn't.
Your point about "no one has done it yet" is irrelevant to what can be done in the future. It's a terrible, low IQ argument.
Systems are degraded in the rain and dark, but there's no proof they are degraded to the point where the solution isn't possible. You are making a leap and trying to present it as a fact.
I think it's clear I'm no Elon/FSD fan boy, but things aren't black and white. Would you accept that Waymo 100% CAN NOT and never will drive on the interstate without a safety driver because they haven't, yet? It's dumb.
Forget the rain and darkness for a moment.
My point that CV/ML isn't safe enough even for unsupervised radiology at this point in time. Driving? Forget it. Perhaps if there are a few major advances in computer vision in the coming years. But not now. Perhaps in two years, five years or ten years. Most likely by the time it happens, if it happens, people will still use physical measurement using active sensors.
I'm not saying it will never happen. I'm talking about where the field is at NOW.
Edit: There are likely no unsolved scientific research problems in the way of Waymo getting to highway speeds.
Last edited:
Knightshade
Well-Known Member
FWIW neither L3 nor L4 are required to be able to drive safely in say heavy rain or snow.
So the fact there's any system impairment in those conditions does not in any way inform what SAE level they're capable of operating at.
You simply ODD them outside in same way current L4 robotaxis ODD out other specific conditions they can't currently operate reliably enough in (like when some brands only operated in cities with great weather, or only during certain times of day, etc).
Only L5 needs to work all of the time a human could.
willow_hiller
Well-Known Member
Sounds like v11 remains in development, for now:
jeewee3000
Active Member
So 11.4.7 could go to wide release within "a week". (make it 3, as usual)
What does this mean to the promised:
- rain sensor update?
- actual smart summon update?
The rain sensor should've been here in July, but I haven't seen it online. The ASS was predicted for September, but I guess we won't talk about that before the rain sensor NN update is here.
Anyone heard updates on this?
What does this mean to the promised:
- rain sensor update?
- actual smart summon update?
The rain sensor should've been here in July, but I haven't seen it online. The ASS was predicted for September, but I guess we won't talk about that before the rain sensor NN update is here.
Anyone heard updates on this?
It's just a bug fix release for 11.4.X, they typically don't deliver anything large. They are trying to fix all the issues and regression in 11.4 from 11.3.6 and haven't been able to. 11.4.6 introduced additional unsafe bugs.
Not disagreeing, but my general take on these releases is a bit more nuanced that that.
Right now with 11.4.4, every time one does an intervention, the car queries the driver as to What Went Wrong. When and if 11.4.7 appears the typical hope is that the new release, with the new code, results in fewer interventions. The Tesla Developers would like to (a) know which interventions are not happening with the new release and (b) if any new ones appear.
The point of this is that, complete re-write or not, a certain amount, probably large, of the logic/NN/whatever in 11.x and earlier is probably going to end up in 12.x. So, the cleaner that code is, the better 12.x will be, re-write or not.
Remember: Our job as FSD-b drivers is to be testers, reporting problems back to the mothership so the product can be improved.
AlanSubie4Life
Efficiency Obsessed Member
Total rewrite in addition to mind-blowing! This is a whole new chapter we have never seen before. Going to be
.What a sad state of affairs when someone who tries to be reasonably balanced about Tesla and FSD can’t help but make fun of Elon. I guess that is what two years (yes, more) of
rosy projections will do!Hopefully it’s better than what we have, I guess. Seems extremely unlikely to yield anything other than incremental improvements. I’m actually surprised that they’ll be able to accomplish the “99%” AI solution without major regressions and I am curious how they will do it. Should be interesting, albeit not very good of course.
Similar threads
