FSD – vår tids største bløff?

[JorEl]

Model S 2018, med alt av utsyr for Full Selv Driving. Til og med betalt/åpnet opp all software.

Er imponert av Robotaxi de siste dagene.
(Ikke like god som "THE ORIGINALE FULL SELV DRIVING" )

Når bilen kjører rundt, uten sjåfør, er dette "Tesla-Fett"
( "Veimo" klarer sikkert minst like bra, men er jo "hardwaremessig" i en annen klasse.)

Stor fremgang. Keep it up.

(..synd Tesla ikke har radar i bilene sine..)

[Motorwatt]

Veimo" klarer sikkert minst like bra, men er jo "hardwaremessig" i en annen klasse.

Kanskje nettopp derfor Tesla i sin nåværende "form" aldri vil nå opp til Weimos klasse?

[daktari]

Fra Reddit:
https://www.reddit.com/r/SelfDrivingCars/s/RqeLKqPD6C

"I did some statistics on the observed failures of FSD robotaxis in Austin

Some initial statistics on observed failures of robotaxi FSD in Austin

TLDR:
We are 95% sure at this point that each Tesla robotaxi can be expected to have an incident of the kinds that have been reported, somewhere between every 2 days to every 8 days.

As follows:

We are now about 3 - 1/2 days into the safety-driver supervised robotaxi test. Reports are that Tesla has deployed a fleet of 10 cars. There have been 11 significant recorded and reported failures of FSD so far:

[https://www.reddit.com/r/SelfDrivingCars/comments/1ljxd63/list\_of\_clips\_showing\_teslas\_robotaxi\_incidents/](https://www.reddit.com/r/SelfDrivingCars/comments/1ljxd63/list_of_clips_showing_teslas_robotaxi_incidents/)

That's enough to do some initial statistics. The confidence centers will be broad because data collection is minimal so far, but we can still derive a failure rate interval and be 95% confident that the actual failure rate is within those bounds.

First, the observed failure rate is simple. 11 failures/10 cars / 3.5 days.  That's a failure rate  of 0.314 failures per day, per car.  That's on average a failure every time a tobotaxi drives a little over 3 days.

But as we said, we have only 3-1/2 days day to so far, so that estimate has a lot of uncertainty associated with it.

This is a time limited observation of discreet events, so there can be modeled as a poisson distribution. We can calculate the confidence interval as follows:

The standard deviation of the number of failures is sqrt(11) = 3.317.
The standard error is SD divided but observation time,  3.317 / 35 car/days = 0.0948
The 95% confidence interval for the rate is: 0.314 +/- 1.96 z * 0.0948
This gives a range of 0.128 to 0.5. Multiplying by 10 cars, we get 1.28 to 5 failures per day.

This is already converging quite rapidly after only 3 1/2 days of data collection. I calculated this after 2 days of data collection, and got a much broader range, but so far three and a half days in these cars are showing a pretty consistent number of failures per day. That cleans the statistical estimates up pretty rapidly.

So based on data to date, Tesla FSD as implemented in the robotaxis, with a fleet of 10 vehicles, can be expected to have somewhere between 1-1/4 to 5 incidents per day, of the kind we have observed so far.

Divide that by 10 to get probabilities per car. We are 95% sure at this point that each Tesla robotaxi can be expected to have an incident of the kind reported, somewhere between every 2 days to every 8 days."

[daktari]

Så kan vi jo lure på hvor hyppig de 5 Ruter-bilene i Groruddalen feiler, de kjører daglig der oppe med kunder og sikkerhetssjåfør.

Hvis Tesla, med feileaten over hadde slått på FSD for 1 million biler, så ville det vært fra 125.000-500.000 hendelser hver dag...