Safety Score

[DCEV]

So everyone gets the engineering version until they cause an accident or some other event that gets them booted out? Sounds like a recipe for a lot of adverse events. For reference, zero accidents have been ascribed to FSD 10 Beta testing in the last year.

https://twitter.com/x/status/1439042334155497474

If you don't click the button, you still have the same access to FSD you did last week.

If you click the button and end up with a low score (which hasn't even been shown that you would), you still have the same access to FSD you did last week.

If you click the button and get a good enough safety score, you may get access to the fresh unproven version(s) of FSD to monitor it.

General releases still go out to all.


Launches don't hurt your score, only braking and cornering G's do.

What they currently have for public FSD is pointless for all practical purposes over what people have by default with Autopilot. certainly nothing worth the $10,000 they have already collected from some people. No turns, no reliable stopping for traffic lights or stop signs.

I'd simply like something useful for the FSD I bought 3+ years ago and I bet many other people who have waited even longer than I.

Launches don't hurt your score, only braking and cornering G's do.

Therein lies the fallacy that acceleration does not cause demerits. I drive my car super carefully. Occasionally when I do launch the car, I don't let the car continue to drive at that speed. I do a quick launch and then immediately slow down (usually with max regen) and I bet many of you do the same to safely and responsibly enjoy your cars.

So as soon as you start slowing down after a launch, guess what, it gets you.

There's no enjoying your car any more and this has nothing to do with your ability to safely monitor FSD Beta.

 

[EchoDelta]

First of all, I think this is an *interesting* approach to determining FSD eligibility. But I do appreciate they are trying something out and it will likely iterate and improve or pivot.

It is especially poignant that Tesla’s own AP on HW 3 doesn’t get good scores driven in smaller rural highways where occasionally there is a tighter turn and a stop light will push too many Gs.

there are car behaviors that will defeat the score: here are my learnings (this is not a post that endorses the system purpose or validity, or teaches you how to game the system, it is just my lessons so that the car itself doesn’t sabotage the score and is not scientific )

1) don’t use AP unless you are in straight or gently bended roads. Sometimes it takes some gradually tightening curves with too much gusto. Got dinged for this.
2) don’t use AP amongst other cars in traffic light roads. It stops too fast too late
3) if your car allows it, temporarily reduce regen. Full regen without feathering decelerates too fast in some situations (eg uphill)
4) turn off Navigate on Autopilot unless you know it knows path, lanes, and roads perfectly. In my case, NoA thought I wanted to be in a certain divided lane and AP decelerated hard as the we passed the divider.
5) set the steering behavior to Chill. Usually I drive our cars in Sport and the softest setting makes me feel I’m driving on jell-o, but the higher g lateral forces are more common with stiffer corrections.

 

[AlanSubie4Life]

I do a quick launch and then immediately slow down (usually with max regen) and I bet many of you do the same to safely and responsibly enjoy your cars

I do not do this. Usually, when I decide I want to floor it, I launch to the speed limit (approximately) and then maintain speed. If I think I am going to have to slow down right away I usually don’t bother launching.

 

[Daniel in SD]

To be clear, I am right, and you just quoted the owners manual confirming that.

THAT said- you're reading too much into what the score says... it says it won't penalize you for stuff on properly-used AP. That doesn't mean it DOES penalize you for improperly turning AP on--- just that if you do so AND something that'd cause a negative score happens- the system will 'count' it, which it wouldn't do if it happened in "proper" AP use.

You are right in this case and most cases.
That's what I was trying to say.

 

[Steve Kempf]

Is your test drive mostly/exclusively using NAV? I don't believe those register.

Yes, I was on autopilot for most of the drive that didn't register yesterday. However, I did a drive this morning, again mostly on autopilot and it did register. So, at least I'm getting a safety score now.

 

[davidsharette]

If you have app version 4.1, logout of the Tesla account on that app and login again.

Thank you!

 

[Sporty]

I think I got “unsafe follow” time for the drive through line at McD’s. I almost want it get the video from the car. It says 7%. The whole drive was 10miles and I wasn’t behind anyone during the drive.

Also got dinged once for “hard stop”. No AP this time. Used the break once to turn off the main road into McDs - .3g is too low in the sense it’s too low for normal situations.

 

[EVNow]

I think I got “unsafe follow” time for the drive through line at McD’s. I almost want it get the video from the car. It says 7%. The whole drive was 10miles and I wasn’t behind anyone during the drive.

Also got dinged once for “hard stop”. No AP this time. Used the break once to turn off the main road into McDs - .3g is too low in the sense it’s too low for normal situations.

Who knew the safety score is also to make sure you don't get junk food

Seriously, they need to filter out low speed FCW, following distance etc.

 

[Knightshade]

It seems like unnecessary semantic pedantry to me to say that penalizing people for events that occur when AP is not appropriately utilized is anything other than…penalizing for inappropriate AP use. I understand it is penalizing for the consequences of that action, not penalizing the action itself.

Not at all.

Lemme make it simple:

Does your score get lowered just for using it inappropriately (which is penalizing you for inappropriate use of AP) or does it only get lowered when a manual-trigger happens AND you're using it inappropriately (meaning it just doesn't ding you for when AP used correctly makes a mistake)

Two very different things.

But I don’t care about that argument. I just want to know when/if AP events are counted in the Safety Score. (And yes it would be good to know the answer about TACC/Autosteer as well. I would assume that if “appropriately utilized,” TACC would mask hard braking and unsafe following. But no idea.)

Well, if they define AP as "TACC and autosteer together" then anything on JUST TACC wouldn't be considered AP, and would count against you for bad behavior.

If they define it as "using any single AP feature" then lots of stuff wouldn't count, including anything on TACC or even hard braking during enhanced summon

What they currently have for public FSD is pointless for all practical purposes over what people have by default with Autopilot. certainly nothing worth the $10,000 they have already collected from some people. No turns, no reliable stopping for traffic lights or stop signs.

I'd simply like something useful for the FSD I bought 3+ years ago

If you bought it 3+ years ago you didn't pay $10,000, you paid $3000 (well, 2-4k most likely depending exactly when)

And for an entirely different set of promised features than the people currently paying 10k.


3 years ago EAP was still a thing. Today, most of the features you get with EAP have moved to the 10k package instead.


Apples and tomatos comparing the features and cost the way you are attempting to.

 

[macmee]

I think I got “unsafe follow” time for the drive through line at McD’s. I almost want it get the video from the car. It says 7%. The whole drive was 10miles and I wasn’t behind anyone during the drive.

Also got dinged once for “hard stop”. No AP this time. Used the break once to turn off the main road into McDs - .3g is too low in the sense it’s too low for normal situations.

I uh, also went to McDonald’s’s last night. Drive through. No unsafe follow though despite a line.

 

[macmee]

I have a feeling you guys are a lot better at whining than driving. Just drive your car like you normally drive your car and see what happens. If you don’t get FSD beta next week so what.

This. I’m just going to do my best and if I don’t get it it’s ok. It’s beta software and when I paid the 10k I knew exactly what I was getting, and what I wasn’t getting. This is an experiment through and through and we’re helping Tesla improve their driving model. They’re also the only company brave enough to release self-driving to thousands of customers and are up against giant regulators in multiple countries and also a press that’s incredibly excited for even just one accident so they can do a big misleading story.

 

[AlanSubie4Life]

does it only get lowered when a manual-trigger happens

Actually you just made it unclear. Are you talking about a user intervention on AP? Because I’d guess all those count, even when AP is being used appropriately on the freeway.

Does your score get lowered just for using it inappropriately (which is penalizing you for inappropriate use of AP)

Yeah as I said we understand that is not happening. There’s not even a category for that, so it clearly does not matter.

But penalizing for actions AP takes while AP is used inappropriately…would be penalizing for those actions. Whether you interpret that as “punishment” for inappropriate AP use, I dunno. Silly semantic argument in my opinion, and mostly I would just like to know when the penalties for vehicle accelerations outside the limits actually apply (when are automation-induced accelerations counted?).

It seems clear enough that automation-induced accelerations caused by appropriate use of AP are not counted. That is all I know about that.

Well, if they define AP as "TACC and autosteer together" then anything on JUST TACC wouldn't be considered AP, and would count against you for bad behavior.

Yeah that is the question.

 

[KArnold]

I think I got “unsafe follow” time for the drive through line at McD’s

Maybe it was more about what you were about to buy? Maybe nothing to do with FSD components?

 

[jjrandorin]

Appreciate all you thirsty people who rushed out to set this up / "push the button". I have FSD but am in no particular hurry to push this button. Maybe in a couple months or so, for me.

Thanks again for all you alpha -2 testers

 

[GWord]

Yup, maybe the solution is to opt out and enjoy a couple of Ludicrous launches and just enjoy driving my car.

But I know this will never come out beta in the foreseeable future. Not even the rain sensing wipers have come out of beta in 3 years. I would like what paid for, knowing what I paid for will be in beta essentially forever.

Just really feeling resentful and sick of this BS "test" criteria that have little to do with safety or your ability to monitor FSD Beta after waiting 3 years and paying for FSD.

If what they were testing had anything to do with safely monitoring being able to monitor FSD Beta I would have ZERO concerns about that. Heck what they monitor has very little to do with actually safely driving the car because it totally lacks any context for the actions.

Alternatively, you could take a day off and, on some interstate near you, rack up 1000 trouble free autopilot miles. Make your denominator so big that a few infractions won’t move the needle.

 

[Knightshade]

Actually you just made it unclear. Are you talking about a user intervention on AP?

No, I'm talking about anything that would ding you if done on entirely manual driving.

So if AP brakes hard but AP is being used inappropriately, as an example of that, you'd get dinged.

Braking hard is a trigger when manually driving. It's ignored if AP is used appropriately and (we can infer) NOT ignored if AP is being used inappropriately because otherwise there'd be no reason to have included the "appropriate" language at all.

But penalizing for actions AP takes while AP is used inappropriately…would be penalizing for those actions.

Right. For those actions, not just "for using AP inappropriately"

If you turn on AP on a road it's not intended to be used on, but drive 100 miles with nothing happening, it has no negative impact on your score. So there's no penalty JUST for using it where not intended.

(honestly there's an argument to be made there SHOULD be... since understanding ODDs is gonna be a pretty important thing for a driver to get correct until someone reaches L5- but that's probably another thread)

Whether you interpret that as “punishment” for inappropriate AP use, I dunno. Silly semantic argument in my opinion, and mostly I would just like to know when the penalties for vehicle accelerations outside the limits actually apply (when are automation-induced accelerations counted?).

I've never experienced any acceleration on AP I'd consider especially aggressive... most folks have the opposite complaint about AP driving, so I'm not sure that'd be an issue either way.

 

[BobDolesGhost]

jk, got off of twitter heh. Someone trying to see if ZERO is possible? Hopefully its photoshopped, wouldn't be hard

 

[AlanSubie4Life]

It seems clear enough that automation-induced accelerations caused by appropriate use of AP are not counted.

And from this in the app, no caveats about appropriate use. So seems like the only inappropriate use in Tesla’s opinion is not paying attention at all and getting forced disengagement.

Others have reported otherwise of course. Probably penalties at other times in their drive?

This matters since it will determine whether (for example) I should use AP on a surface street with many traffic lights (where the risk of a user hard braking event due to yellow lights is high), where I would normally rarely use it, for optimizing my score.

As above, also can rack up those AP miles following another vehicle to get a large denominator…others here have claimed it does not work that way but my assumption is they are being penalized for events while under manual control. In addition to the denominator effects, this would be especially valuable in rush hour or busy traffic on the freeway over 50mph where cut-ins would destroy your score if in manual mode.

I think if I had used AP, nearly all my demerits so far on unsafe following would have been masked. They occur when people change into the lane in front, and it is not possible to open up 1 second instantaneously.

 

[kbM3]

Lol, I have Tesla insurance and got a 97 today, with my only ding being “hard braking” when I never once hit the brake pedal. The only time it could have been “hard braking” was when I was coasting down the large hill I live on top of and have gotten good at timing properly to coast to a stop at the light that’s at the bottom without once hitting the brake. This seems to be confirmed because I was at a 95 when i checked after getting to the store after driving down the hill, and got back up to a 97 when I checked after I finished my errands and I got back home. I wonder if I should be calling to complain about how they’ve been calculating my premium?

It seems plausible that acceleration is measured by the accelerometer not the derivative of wheel speed. This method makes sense because braking distance will be increased when going down a hill.
A 10% grade (steep hill) is 5.7 degrees.
sin(5.7) * 1g = 0.1g
If regen is 0.2g then the extra 0.1g could push you over the 0.3g limit!

Seems unlikely downhill was the problem

Hard braking is defined as losing about 6 mph of speed every second.

If you’re going UPHILL and let off the accelerator, you’re going to decelerate more because gravity and being at higher risk of slowing more than 6 mph/s.

Conversely letting off the accelerator while going downhill results in less deceleration.

You can even ACCELERATE with your foot off the accelerator altogether on a steep enough decline.

 

[⚡️ELECTROMAN⚡️]

I wonder if McDonald's patrons have worse safety records than average