dduffey
Member
In the App, click in your profile, scroll to the bottom and look at the App version. It needs to be at least 4.1.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Safety Score
- Thread starter ScottNY
- Start date
That worked. I deleted the app and re downloaded it. I didn’t have 4.1. I do now and my safety score is now there. Who’s bettah than you!
bjrosen
Member
How have you managed a 99? I'm at 96 but getting higher than that seems impossible. It regards any use of the brake peddle, no matter how light, as hard braking. I'm doing one peddle almost all of the time but there is the occasional instance when the car isn't slowing down enough by itself and you have to tap the brake, this is especially true when the charge level is high and the car hasn't had time to warm the battery. Yesterday on the start of a long trip I had to use the peddle at a stop sign near my house, it was a light tap but I got dinged on my score. Curiously on my way home I did have a hard stop when someone in front of me stopped suddenly, that stop didn't show up on my safety score. I also had one instance of following too close which I know never happened, I normally keep large distances between me and the next car and yesterday I was especially vigilante. There were no warnings and I have the warning level set at lower distance than they are supposedly using for this score.
dduffey
Member
Being in MA, it's going to be difficult.
I had a 93 on the first day because I didn't realize that keeping a follow distance over 3 seconds doesn't help my score.
I'm luck to be in Texas and flexible with my schedule, so I can do my driving when it is calmer and fewer traffic.
It's really measuring what environments are the safest to roll out to, and the roads and other drivers are just as much of a factor as the driver.
I had a 93 on the first day because I didn't realize that keeping a follow distance over 3 seconds doesn't help my score.
I'm luck to be in Texas and flexible with my schedule, so I can do my driving when it is calmer and fewer traffic.
It's really measuring what environments are the safest to roll out to, and the roads and other drivers are just as much of a factor as the driver.
I had to valet my car last night and the valet got me some hard braking and hard turns, but thankfully no FCW. I forgot to put the car in valet mode, I wonder if that would have kept them from dinging me. I did manage to get back to 100 for the day, but had to drive around in a roundabout for 4 minutes, lol
bjrosen
Member
I don't commute, I only do day trips on Saturday's so I should be in the best possible conditions. Yesterday we went to Manchester VT, the whole 260 mile trip was on rural roads with very little traffic, did something similar the Saturday before, we went to Western MA so mostly rural roads a little bit of highway. You are right that if I was commuting in MA it would look terrible, Route 128 is essentially a parking lot, bumper to bumper with speeds dropping to 2MPH (that's not hyperbole, that for real).
This is my 4th Tesla and regularly use AP. This was my first ding on my first day using FSD on the Merritt Parkway in Connecticut. This Sunday morning was manually driving Merritt at 8 AM. Some guy cut closely in front of me and sped away. Got dinged again for “following too close”. I think there is a glitch in FSD 10.1 that will hopefully be fixed in 10.2z
AlanSubie4Life
Efficiency Obsessed Member
What’s the glitch? This seems correct. I’m sure there will be changes in 10.2 though!
Overall it seems likely that no one is dinged when using AP. Still no firm evidence…but I think most people blaming AP for dings just were confused about how the scoring works - using AP can certainly lead to much worse scores, as discussed - but not due to AP.
Last edited:
Teefal
Member
Reading through this thread from the start. Here are some thoughts that might be helpful.
I have experience training neural nets for professional sport game video annotation, and have been a software engineer/manager for decades. I am certain of a few things:
* The "safety score" is a brand name that Tesla unveiled to gain mindshare on the concept for later uses related to insurance breaks, much like Geico and Progressive, only MUCH better (eventually)
* As it stands now, the safety score had nothing to do with safety. That word is tripping people up because it creates a false correlation between what their common sense of the word safety entails and what the actual goals of the fsd beta score actually are.
* Tesla's goal NEEDS to be to find drivers who are just a little bit better than their current nets. This is the only way to incrementally improve them. Otherwise they have to throw out a ton of training data that's too far ahead (race car drivers). It's not about safety. It's about picking people who demonstrate an ability to patiently drive just a bit better than their nets. (It's too hard to distinguish race car behavior from reckless behavior.)
* Beyond this implicit matchmaking, they need to find people who are abnormally patient, attentive, and focused on pushing the cars just a bit further (like AI Driver), who have the patience and focus of a QA engineer. The most useful data will come from these drivers. The ability to game the system to get a score is actually a benefit in finding people with this abnormal patience, attentiveness, and willingness to risk the wrath of irate drivers. These skills are necessary in test driving the beta.
* Yes, Tesla wants "safe" drivers to avoid bad press and regulatory scrutiny. But thats kind of a gimme. If you screen out impatient/inattentive drivers, the likelihood of bad headlines goes down.
* This is NOT an FSD preview. They are ramping up the TRAINING, and it will suck spectacularly during many edge cases that haven't been handled yet. The more people veer from a true QA mindset, the more frustrated they will be, and the less useful their data will be.
* If you are angry about the arbitrary metrics and unfair dings, if you are going crazy that the safety score isn't allowing you to drive the car like the Tesla you bought ... guess what, you are probably not the best candidate to teach the car the NEXT phase of its evolution. Yeah, you might be Mario Andretti with reaction times like a Ninja. Yeah, you probably haven't gotten into an accident for 20 years. But its NOT about safety or skill. Just as Agassi wouldn't be the best tennis coach for a 12 year old, you are not a good coach for baby driver beta.
Just my 10 cents.
I have experience training neural nets for professional sport game video annotation, and have been a software engineer/manager for decades. I am certain of a few things:
* The "safety score" is a brand name that Tesla unveiled to gain mindshare on the concept for later uses related to insurance breaks, much like Geico and Progressive, only MUCH better (eventually)
* As it stands now, the safety score had nothing to do with safety. That word is tripping people up because it creates a false correlation between what their common sense of the word safety entails and what the actual goals of the fsd beta score actually are.
* Tesla's goal NEEDS to be to find drivers who are just a little bit better than their current nets. This is the only way to incrementally improve them. Otherwise they have to throw out a ton of training data that's too far ahead (race car drivers). It's not about safety. It's about picking people who demonstrate an ability to patiently drive just a bit better than their nets. (It's too hard to distinguish race car behavior from reckless behavior.)
* Beyond this implicit matchmaking, they need to find people who are abnormally patient, attentive, and focused on pushing the cars just a bit further (like AI Driver), who have the patience and focus of a QA engineer. The most useful data will come from these drivers. The ability to game the system to get a score is actually a benefit in finding people with this abnormal patience, attentiveness, and willingness to risk the wrath of irate drivers. These skills are necessary in test driving the beta.
* Yes, Tesla wants "safe" drivers to avoid bad press and regulatory scrutiny. But thats kind of a gimme. If you screen out impatient/inattentive drivers, the likelihood of bad headlines goes down.
* This is NOT an FSD preview. They are ramping up the TRAINING, and it will suck spectacularly during many edge cases that haven't been handled yet. The more people veer from a true QA mindset, the more frustrated they will be, and the less useful their data will be.
* If you are angry about the arbitrary metrics and unfair dings, if you are going crazy that the safety score isn't allowing you to drive the car like the Tesla you bought ... guess what, you are probably not the best candidate to teach the car the NEXT phase of its evolution. Yeah, you might be Mario Andretti with reaction times like a Ninja. Yeah, you probably haven't gotten into an accident for 20 years. But its NOT about safety or skill. Just as Agassi wouldn't be the best tennis coach for a 12 year old, you are not a good coach for baby driver beta.
Just my 10 cents.
Last edited:
I believe that Tesla sets to "medium" when it comes to their scoring regardless of the setting done in each car.
I believe that Tesla automatically sets collision warning to "medium" for all cars regardless of what the individual settings are within each car.
dduffey
Member
I am 300+ miles at 100. Today made a short trip and got a "hard braking". Pretty sure it was when I pulled into a parking spot and braked. dang! Still at 100, but now 98 for the day.
Now gotta go out and do some light braking to bring up the score for today.
Now gotta go out and do some light braking to bring up the score for today.
EVNow
Well-Known Member
IMO, the "safety score" is what it is. They used some kind of ML to find a way to score people's driving using the current data they have. The score as they have calculated correlates with accidents (i.e. higher the score lower the chance of accidents).
But - because they don't filter out edge cases - what has happened in practice is you need to drive very patiently to rack up your score. Its just a coincidence rather than the design.
PS : Does anyone know if the way they are scoring (i.e. the formula) is similar to what other insurance providers or some research papers use ?
I was on autopilot today and my car took too long to start braking when the cars ahead were slowing down and once it finally started braking, I got a collision warning. I remained an autopilot the entire time. I didn’t get dinged maybe because either (1) my car is set to early and the event would not have triggered a ‘medium’ collision warning or (2) like some of the other factors, collision warnings on autopilot do not count against you. I believe point 2 is correct, but I have no explicitly seen that stated and before today didn’t even realize you could get collision warnings while on autopilot.
mackgriggs
Member
I mean, its similar. Last year I switched to Nationwide and they have a safe driving program that tracks you for 30 days, using your phone. It only counted hard braking and hard acceleration (not hard turning), since it only used the phone's g-force sensors, though it also tracked your idle time (meaning, it assumes drivers in stop and go traffic get in more accidents, or, if you were always doing city driving instead of freeway, you would have more idle time, and since city driving is more dangerous than freeway, would count against you). It also meant...if you were driving and your phone accidentally dropped on the floor, it would ding you.
SmartRide Video – Nationwide
Watch this video to learn how SmartRide from Nationwide can help you save up to 40% on your car insurance rates using your mobile device.
Nationwide does show you your numbers and miles driven and shows your current discount value and whether you have green/yellow/red range zones.
Just like this Tesla scoring...it made driving SUCK. You would have to try and game the system, just like this one. With the nationwide one, you could turn on/off the program at anytime by starting/stopping trips. It even got to the point (since I do all city driving), that I would only turn it on when I knew I didn't have a lot of stoplights, or...i would stop the trip at a long stop light, and then start a new trip.
I had just bought the MY too when I switched...so it was disappointing to have to drive my essentially new sportscar like a grandma for a month.
AlanSubie4Life
Efficiency Obsessed Member
I think preemptively, we're going to have to start discussing fractional scores for this coming Friday/Saturday.
It seems perfectly legitimate to have Tesla discern amongst the people with "100" both based on distance (I don't think the minimum mileage should be any more than 100-200 miles - I don't think huge distances necessarily indicate any sort of aptitude - just that these people have long distances able to be conducted in uncomplicated conditions, which isn't really what they're looking for), and more importantly, based on fractional score.
Really, Tesla should do an app update to let people see their actual scores without rounding, otherwise there could be a lot of upset if there are more than 1000 people with 100 scores displaying next Friday evening. Seems very predictable (I got a 100, I didn't get in, waah-waah,
, Tesla is lying, etc.)!If using the true unrounded daily values, mileage weighted, I have a 99.3.
Rounding the individual days first, I have a 99.6.
Overall, in the app, I have a 100.
This is with 365 miles of suburban and freeway driving.
My goal is to get to 99.6 on the true unrounded daily value weighted average, and a 99.8 on the rounded individual days weighted average result. It's possible that that will be good enough. It'll be hard though - requires 300 miles of nearly perfect driving, and while my scores have improved, on average, true perfection is still sometimes not fully within my control.
And of course there is always the very small risk of an FCW - I got one last night, but on the Early setting (it was ridiculous), and it did not show in the app. A single one of those would put me completely out of the running for the first group unless I drive 500 miles (!!!!) on that same day to fully make up for it (alternatively 166 miles is the optimal distance to drive on that same day, to bring the daily rounded score up to 99 (around 98.6), which would mean a total distance of about 330 miles to drive to fix it up). And I can't use AP around the neighborhood, where these are most likely to occur due to parked vehicles. So it's definitely a dangerous situation for my score!
Anyway, I think right now it's quite possible my 100 score is just not good enough to make the initial cut. (Excluding my first day of driving, I'd be at my target 99.6/99.8 - really should have worked to polish up that first day above 98.51 - but did not fully understand at that point how easy it was to improve individual daily scores - and unfortunately it was a high mileage day coincidentally.)
If I thought it was good enough, I'd probably stop driving, or at least do extremely well-controlled quiet evening drives only. I guess I could opt-out selectively but not sure how that will be viewed and whether it changes queue position (seems a bit less egregious than resetting the computer after a screw-up, because at least it shows self-awareness, without the ability to truly cherry pick drives).
Last edited:
Similar threads
