EVNow
Well-Known Member
I dread having to drive myself because of rain, for eg. So, clearly FSD is a useful ADAS.
You really need to stop being the cup is 1/10th empty guy !
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
You really need to stop being the cup is 1/10th empty guy !
AlanSubie4Life
Efficiency Obsessed Member
I am not sure I follow. Is FSD making you safer in the rain? Does it work in the rain?
AlanSubie4Life
Efficiency Obsessed Member
100% agreed with that. Useful generalized L2 is much harder than L4. (It has to be everything L4 is, and then additionally needs to be spectacular enough to make the driver comfortable with accepting liability and willing to be happy with 100% monitoring the task while not actually doing it.)
100% agreed with that. Useful generalized L2 is much harder than L4. (It has to be everything L4 is, and then additionally needs to be spectacular enough to make the driver comfortable with accepting liability and willing to be happy with 100% monitoring the task while not actually doing it.)
EVNow
Well-Known Member
I miss it when it doesn't work ergo it's been useful when I can use it.
AlanSubie4Life
Efficiency Obsessed Member
Oops. Was going to consolidate posts, but then got interrupted in the middle. Ah well.
I'm looking for improvements primarily on restricted access highways so good chance the over-all disengagement rate won't improve significantly but the limited access highway "intervention" rate could decrease a lot for most drivers.
Agree with this. My complaint has been that FSD in moderate/heavy rain disengages too much but when it does work especially at night on roads I've never driven before it's very helpful. Last Thursday night in light rain FSD because of a traffic accident rerouted me from the highway to windy secondary roads. FSD was extremely helpful.
jebinc
Well-Known Member
You meant, “9/10ths” empty (or 1/10th full), right?
$15,000 for a not upgradable maybe L2 system; thank you please!
EVNow
Well-Known Member
That 1/10 comment is just a take on the ULT bets - nothing really to do with FSD quality....You meant, “9/10ths” empty (or 1/10th full), right?
$15,000 for a not upgradable maybe L2 system; thank you please!
FSDs excess processing delays (FSDb time is not not real-time), design crutches (30ft crawls, myopia, jerky throttle/brake/steering), and generally being out to lunch likely indicates the team is doing what they can to get the software on the road while knowing future designs will bring more hardware processing and possibly more capable sensors.
Tesla's original hardware design with 2 core 'redundancy' shows little homework was done upfront as current hardware is likely 4x to 8x insufficient for the current design. Add more capable sensors, a goal of fewer 'corner cases,' real-world nimbleness, and I would guess hardware needs grow closer to 10x. All out of my back-side assuming TSLA wants to keep a goal of hardware redundancy (I bet they don't).
Tesla's original hardware design with 2 core 'redundancy' shows little homework was done upfront as current hardware is likely 4x to 8x insufficient for the current design. Add more capable sensors, a goal of fewer 'corner cases,' real-world nimbleness, and I would guess hardware needs grow closer to 10x. All out of my back-side assuming TSLA wants to keep a goal of hardware redundancy (I bet they don't).
Last edited:
Knightshade
Well-Known Member
The HW already is not redundant (and never really was in actual use).
As to how much "more" processing power they need.... nobody knows. Including Tesla.
Nobody WILL know until they get a working system.
As to how much "more" processing power they need.... nobody knows. Including Tesla.
Nobody WILL know until they get a working system.
So is FSD Feature Complete now? Or is ELon just shocked (for like, the 10th time)?
Tesla CEO Elon Musk said during the Q4 2021 Earnings Call that he is confident the company's Full Self-Driving suite will be finished by 2022. Musk was extremely confident in his prediction, stating he will be “shocked” if Tesla does not complete FSD by the end of the year.
Tesla CEO Elon Musk said during the Q4 2021 Earnings Call that he is confident the company's Full Self-Driving suite will be finished by 2022. Musk was extremely confident in his prediction, stating he will be “shocked” if Tesla does not complete FSD by the end of the year.
jebinc
Well-Known Member
The Charlatan will just redefine FSD again to match the current city streets abomination that was released during the holiday’s.
Knightshade
Well-Known Member
FSDs definition has matched the definition of city streets since March 2019. Like, that was literally the definition and name of the only undelivered part of FSD since then.
It's only in the FSDb that that feature was actually available on cars.
This morning I checked to be sure and only the one route offered. It also “guessed” wrong and tried to route through the construction site with it blocked off so I had to disengage. So it is a disengage if if choses the alternate route and also if the choses the main route and it is blocked.
Just for kicks here it is going the opposite direction making a right turn from 16th St onto West Peachtree St and trying to cross 5 lanes to make a right turn on Arts Center Way. It actually has more distance this way than the right turn off Arts Center Way to left on 17th St. There was almost no traffic this morning so I let it play out and 100% can’t do it and misses by 2 lanes.
AlanSubie4Life
Efficiency Obsessed Member
You meant, “9/10ths” empty (or 1/10th full), right?
Yep, for some, around 90% is good enough (ULTs sadly have STILL not made it to 90% unless we missed something in Chuck’s recent testing - he seems very sad to not have V11 and may be silently boycotting, hard to say).
For me: I believe FSD is nearly 9/10ths of the way there, and that is nowhere near good enough (results in 2-3 disengagements per mile with the way issues are currently distributed).
May the resolution for 2023 be to get the first 9! The second 9 would be great but I think is way too much to hope for. Probably need 2-3 9s for it to be useful (if a bit dangerous and needing to be monitored rigorously).
My main hope for 2023 is to get awesome & robust driver monitoring. This is the way. It would be so awesome.
powertoold
Active Member
I'm pretty sure Tesla has a simulated or actual HW4-equivalent system running with larger models.
AlanSubie4Life
Efficiency Obsessed Member
And I am pretty sure we would know about it if they had solved all the problems with that hardware.
Honestly for me it seems like if they could figure out how to slow down and how to speed up, and allow me to control lane changes, that would solve 90% of my problems and be usable in a broad range of suburban scenarios. And that seems like it should not require new hardware.
Not that that would meet the objectives - but would make much more usable.
Of course, not being usable may well be a feature.
Currently many limitations due to software, and probably many limitations due to hardware.
powertoold
Active Member
I'm assuming your slowing going issues are similar to the prior ones.
I think most of all our annoyances stems from the intersections / road geometry / semantics training. During AI Day 2, Tesla said they trained on over a million intersections, but I think this NN model has to be distilled quite a bit to fit into the HW3 at runtime. Either this is the case, or Tesla is still slowly auto-labeling more and more intersections (which takes time), so the improvement will be linear over time...
Similar threads
- Article
