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

Wiki MASTER THREAD: Actual FSD Beta downloads and experiences

This site may earn commission on affiliate links.
Seems unlikely that the majority of them would be, Tesla requests opt-in before collecting the score. Why would they opt in to having the score collected with no benefit to them?
Only if it was part of the insurance thing. The law only prohibits using driving stats to calculate insurance (until it can be shown to be not discriminatory) - not collecting stats !
 
  • Informative
Reactions: Phlier
Only if it was part of the insurance thing. The law only prohibits using driving stats to calculate insurance (until it can be shown to be not discriminatory) - not collecting stats !
What I meant was that I should imagine Tesla is unlikely to collect the data without opt-in for that scenario if they won't collect it without opt-in for the FSD beta testing scenario. Perhaps you understand that and mean that neither of us know much about Tesla insurance, so Tesla could require opt-in to participate in insurance as well. I agree with that assessment, but @MP3Mike appears to have indicated otherwise, and he may know.
 
  • Like
Reactions: impastu and EVNow
This says different.


They collect anonymous data.
But does the anonymous data include a safety score? It's not on that list, and they couldn't calculate one after the fact if those statistics were anonymized separately before being collected.
 
  • Like
Reactions: impastu
They collect anonymous data.
Correct, they collect anonymous data from the entire fleet. (Other than people that have turned off data sharing.) That is how they came up with the Safety Score calculation.

But the entire fleet isn't using the Safety Score. Only the people that have opted in to it by clicking the FSD beta request button or signing up for Tesla Insurance in Texas are using it.
 
  • Like
Reactions: impastu
Somewhat off topic, but did not want to start another thread. Debating whether to opt in. I’m on software that includes opt in and own FSD. Does that mean that the AP functions I have now (lane keeping, lane change, NOA…) are using FSD code and cameras only, or are they still using old radar code?
Unless you've installed 2021.32.25 or 2021.32.26, you are on the old radar code.
 
You're driving along and some typhlotic driver pulls right out in front of you. How does it make you a bad driver if you jump on your brakes in order to avoid a collision that would have been caused entirely by the other driver?
Exactly!!

Not knowing the context of every "hard braking," "aggressive turning," etc. completely destroys the accuracy and usefulness of these as metrics to determine safety. And don't even get my started on the flawed logic of the Forward Collision Warning, which is even worse in FSD beta than it has been, ever.

There's a spot I drive through daily that has given me FCW's every time I've driven this road since downloading the beta. Never had a problem with it before. It's a gentle curving road that has a traffic light and a dedicated right turn lane. If you're in the straight lane, and there's a car in the right turn lane waiting to turn, you'll get nailed with an FCW every time you go through that intersection.

Now sure, there's some argument to be made for patterns... if a guy constantly gets dinged for hard braking and aggressive turning, there might be something to be said for the safety of this driver. But, where is this driver located? In many locals, the average braking and turn rates are higher than in other places. So once again, context matters.

I'm sure trying to get accurate metrics to measure driver safety is a difficult problem to solve, and IMO, Tesla isn't there yet.

My wife hasn't been in a traffic accident for 30+ years, but I know that if she were to drive under Tesla's Safety Score program, she'd be lucky to hit 50; her safe driving habits are simply different; but different does not mean unsafe.

But Bill... honestly, though... "typhlotic?" You must be an old guy like me. ;)
 
But does the anonymous data include a safety score? It's not on that list, and they couldn't calculate one after the fact if those statistics were anonymized separately before being collected.
The way most companies anonymize data allows individual cars to be tracked. But you won't know which car it is. Anonymized data is retained at granular level and aggregated when needed. For eg. they could use one-way encryption of the vin. So, they get a consistent key to track the data coming from a particular vin - but noone can figure out what the actual vin is.

Just linking the first relevant search I found.


ps : Note that privacy rules in US are non-existent / hazy (expect to change with the new CA law). So, most companies now follow EU GDPR rules - which allow for anonymized data collection and usage, without any opt-ins.

While the GDPR is strict, it permits companies to collect anonymized data without consent, use it for any purpose, and store it for an indefinite time—as long as companies remove all identifiers from the data.
 
Last edited:
Got 10.2 last night. My main concern is that it's driving way too close to the center of the wider unmarked 2-way streets in my neighborhood - anyone else feel that way? Almost a car width too far to the left and closer than I feel comfortable when there's an oncoming car.

Also probably the first time I feel an HUD would be helpful. It's hard to guess whether it's about to turn too closely to a parked car without checking the path planner, which is too far out of the line of sight when the car is making the maneuver already - specifically the case of a right-hand turn where there is a car parked at the corner of the street you are about to turn onto. Maybe it's just a matter of establishing trust by using it more.

Finally, mapped but hard-to-see turns (e.g. into a store parking lot that's mapped correctly on the nav) are a bit chaotic with the steering wheel movements as well as the overall speed, though I was impressed that it was able to find the right entrance in the first place. I ended up disengaging right after making the turn because it was moving a bit too quickly onto the sloped entrance.
 
Got 10.2 last night. My main concern is that it's driving way too close to the center of the wider unmarked 2-way streets in my neighborhood - anyone else feel that way? Almost a car width too far to the left and closer than I feel comfortable when there's an oncoming car.

Also probably the first time I feel an HUD would be helpful. It's hard to guess whether it's about to turn too closely to a parked car without checking the path planner, which is too far out of the line of sight when the car is making the maneuver already - specifically the case of a right-hand turn where there is a car parked at the corner of the street you are about to turn onto. Maybe it's just a matter of establishing trust by using it more.

Finally, mapped but hard-to-see turns (e.g. into a store parking lot that's mapped correctly on the nav) are a bit chaotic with the steering wheel movements as well as the overall speed, though I was impressed that it was able to find the right entrance in the first place. I ended up disengaging right after making the turn because it was moving a bit too quickly onto the sloped entrance.
Yeah I've seen the centering behavior on unmarked roads as well as a number of other folks. I feel like its some recent regression because I remember watching a lot of FSD videos in the past where it didn't do this.