They seem to have 4 separate 4G modules with 4 different operators as fallback. But they say they use only one in active mode. So I guess it’s possible. Thanks for that.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD discussion
- Thread starter Buckminster
- Start date
Didn't see any improvements for my usual disengagements:
1. unprotected turns, in mid heavy-traffic
2.indecisive when single lane splits into two
Once FSD stops slowing down at stop sign(which is usually 3-5feet ahead of actual intersection), and creeps and stops all the way to the intersection line, I think unprotected turn behavior will improve greatly.
But some things are like human behavior.
#Was at traffic light, like 4th car from lights. Suddenly my car moved forward and I was like shocked and was about to disengage. Then realized that FSD created a slightly larger gap behind my car, so that the car on other side of the road had enough space to cross and enter LIDL
#Also sometimes it doesn't turn to right most lane, but the middle lane because right most lane has more cars.
..things I would normally do.
# At stop lights if there is a large truck blocking the lights, it just follows the truck until it can see the lights
there was a video of FSD following human turn signals, so lets see how that progress goes. though it is kinda of scary, as anyone could show a stop sigbn and cause the car to stop. Laws will likely need to be written for this.
+ parking works well- but seems it likes to reverse into the parking spots
1. unprotected turns, in mid heavy-traffic
2.indecisive when single lane splits into two
Once FSD stops slowing down at stop sign(which is usually 3-5feet ahead of actual intersection), and creeps and stops all the way to the intersection line, I think unprotected turn behavior will improve greatly.
But some things are like human behavior.
#Was at traffic light, like 4th car from lights. Suddenly my car moved forward and I was like shocked and was about to disengage. Then realized that FSD created a slightly larger gap behind my car, so that the car on other side of the road had enough space to cross and enter LIDL
#Also sometimes it doesn't turn to right most lane, but the middle lane because right most lane has more cars.
..things I would normally do.
# At stop lights if there is a large truck blocking the lights, it just follows the truck until it can see the lights
there was a video of FSD following human turn signals, so lets see how that progress goes. though it is kinda of scary, as anyone could show a stop sigbn and cause the car to stop. Laws will likely need to be written for this.
+ parking works well- but seems it likes to reverse into the parking spots
Last edited:
Doggydogworld
Active Member
Phantom Auto also does teledriving, though it seems they've pivoted more to industrial markets. I think one of those little food delivery bot companies (maybe Starship?) uses remote drivers based in a low income country. As @Knightshade says, though, Waymo and Cruise remote monitors do not actually drive the car, just help with route selection and such when a car is unsure how to proceed. Sometimes the remote monitors screw up -- in January one mistakenly told a Waymo to proceed through a red light.
willow_hiller
Well-Known Member
As many of us guessed, the take rare is much higher than 2% and Yipit's data was rubbish:
I'd wait until we get actual data from Tesla before celebrating this. Technically 4% is double what that survey said which would count as "much higher than 2%", but at the same time I don't think anyone would celebrate it.
Saying that, if Tesla is able to get 100x improvement on DE frequency, it would put it around 1000km on average per disengagement and suggest they're on the right track towards achieving unsupervised FSD and, by extension, RT. Judging by the data available, they couldn't release RT now.
Last edited:
willow_hiller
Well-Known Member
Did you say the same thing to the bears celebrating the poorly modeled Yipit data?
No, as I was still waiting on clarifications. Again, we don't know how good the Yipit data is. My personal take is that the number isn't impressive, it's the same story as it was with the Model 2 Gen3. Technically the original news gets one piece of the puzzle wrong and you can call it a lie. In this case, maybe it's 4% or even 6% (300% bigger than the estimate), but way below even the most pessimistic expectations.
Given Elon's style of posting, as an investor, you can look for some pretty useful clues. If the actual numbers are impressive, they'll be made public. See 12.4 touted improvement of 5-10x, then 12.5 with similar numbers. Otherwise it will be countered on details, but make it seem like the whole story is wrong. In the long term it doesn't matter, every stock is evaluated on projected future earnings, the more deadlines are being missed, the more circumspect the market will look at "pie in the sky" projections for the future.
Changing lanes over the solid white is illegal. My wife actually got a ticket for that in Philadelphia, when she realized the long line of cars in the left lane was for an exit up ahead. Someone let her in, but it cost her $125. Pure revenue stream for excitable PDs. Programmed FSD will never do that.
Seems Tesla has paused 2024.14.6 for unknown reasons.
Tesla Update 2024.14.6 Statistics & Roll Out
Also, speaking of real time view we discussed here last week, coming soon to an app near you:
willow_hiller
Well-Known Member
There are some small, but significant changes to the ways in which those two features will work in 12.4.
On the vision-based attention monitoring, you no longer need to be holding the steering wheel as long as:
- the camera is not occluded
- there is sufficient lighting
- the driver is looking forward
- the driver is not wearing sunglasses
- the driver is not wearing a low-brim hat or another object that covers their eyes
And on Autopilot suspension, strikes will expire after 7 days, unless another strike is earned during that week at which point the 7 day timer starts over again.
BitJam
Active Member
How do you earn a strike while you're on autopilot suspension?
What he meant, or based on the release notes, if you have less than 5 strikes you can continue to drive in FSDS. And if you behave well and get no strikes for one week, that strike count that you already accumulated will decrease by one.
So if you get your 3rd strike today, and you drive for one week without getting any additional strike then that strike count will be reduced from 3 to 2.
I can't remember the last time I drove during the day without sunglasses. But I'm OK with nagging the wheel.
fasteddie7
Active Member
Maybe because they paused the rollout of the previous version so it’ll be new to most?
This is another wonderfully vague reply by Elon. Remember how he said the Reuters article about the $25k compact being canceled was a lie?
Well sure, it was "kind of" a lie, in that the article implied that there would be no cheaper model. But it was correct, strictly speaking, in that the $25k model was dead. They did not include that Tesla had decided to accelerate the launch of a cheaper model by doing it a bit differently, and maybe at a slightly higher price.
We've seen this quite a few times by now. A number of claims are made, and Elon makes a blanket "they are lying" statement, but then it turns out it was maybe only partially a lie or maybe just a little bit misleading, but the gist of it was mostly correct.
So how much faith can we put in a vague reply like this from Elon?
willow_hiller
Well-Known Member
I put more faith in a vague reply than poorly documented data. I made a couple of posts about it on the main investor thread, you can see my thoughts on the modeling that likely went into the Yipit data here: Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
tl;dr is that all Yipit likely knows is they obtained the credit card data of 3,500 people who had ever made a payment to Tesla, and then after an arbitrary 45 days found that 2% of them had subscribed to FSD. But they have no way of knowing who in that pool actually had the trial. There are also glaring errors in the data, like showing the first conversion on 4/12 when the trial had barely been running for 2 weeks. Who subscribes to a service 2 weeks into a 1 month free trial?
Similar threads
