Safety Score: parameter dependence

[Michael_Vinson]

Hi everyone. Apologies if this has already been discussed; I'll keep this short in case it has. As of last night, my score was 80, and I broke it down into its components so I could assess the marginal contribution of each of the five factors. See the plot, below. For the less mathematically inclined, this plot suggests that working on "Forward Collision" will have the largest effect on my score, whereas "Unsafe Following Time" will have almost none at all. "Hard Braking" will very quickly lower the score, gotta watch that one. All the better to game the system, haha.

Michael Vinson

 

[Dan D.]

Good plot! Hard Braking is very significant. Just 1.4% knocked off 5 of your safety points.

It's the combination of the factors that really gets you, having numbers in more than one category makes each factor worse. It's not the sum of the factors, each one multiplies the effect of the other ones.

For example, Hard Braking at 1.4% by itself would drop your score from 100 to 97.2 or -2.8 points
but add in your other factors and now Hard Braking at 1.4% changes your score by -5.4 points

 

[ATPMSD]

Safety Score Beta | Tesla Support

The Safety Score Beta is an assessment of your driving behavior based on several metrics called Safety Factors. These are combined to estimate the likelihood that your driving could result in a future collision. We combine your daily Safety Scores (up to 30 days) to calculate the aggregated...

www.tesla.com

 

[Michael_Vinson]

Good plot! Hard Braking is very significant. Just 1.4% knocked off 5 of your safety points.

It's the combination of the factors that really gets you, having numbers in more than one category makes each factor worse. It's not the sum of the factors, each one multiplies the effect of the other ones.

For example, Hard Braking at 1.4% by itself would drop your score from 100 to 97.2 or -2.8 points
but add in your other factors and now Hard Braking at 1.4% changes your score by -5.4 points

Excellent point that they combine multiplicatively, Dan. Absolutely true. I'm still too much of a noob to be allowed to "Like" a post but consider this reply my Like!

The app shows a gray line for "Fleet Median" values of each parameter. Eyeballing them I get about:
Forward Collision Warnings: 11
Hard Braking: 1.5
Aggressive Turning: 3
Unsafe Following Time: 16
Autopilot Disengagement: 0

These (approximate) median scores give a safety score of 92, for whatever that's worth. If anyone has better values for the fleet medians, I'd love to know.

Cheers,
Michael

Proud owner of 2021 Model Y long range

 

[BCNY]

So if there is something completely unexpected in front of me and I hard brake and swerve and successfully avoid being in an accident my score is lowered???

At the same time if another Tesla driver crosses 5 lanes without even using the turn signal, he is a safer driver?

These parameters are stupid. They are there probably because they are easy to implement and they do check a box of "measuring something"

 

[tekdemon]

So if there is something completely unexpected in front of me and I hard brake and swerve and successfully avoid being in an accident my score is lowered???

At the same time if another Tesla driver crosses 5 lanes without even using the turn signal, he is a safer driver?

These parameters are stupid. They are there probably because they are easy to implement and they do check a box of "measuring something"

It’s averaged over all the miles you drive so if that happens once it’s not a big deal. You just can’t consistently be doing worse than everyone else on these parameters.

If someone really regularly swerves over five lanes without signaling they probably do a lot of other garbage driving.

I get that if you’re in nyc it can be common to be bumper to bumper but you basically need to use AP during all those periods where its likely to ding you and then manually drive with a safe following distance when you’re on a nice relaxed highway.

 

[NJM3]

So if there is something completely unexpected in front of me and I hard brake and swerve and successfully avoid being in an accident my score is lowered???

At the same time if another Tesla driver crosses 5 lanes without even using the turn signal, he is a safer driver?

These parameters are stupid. They are there probably because they are easy to implement and they do check a box of "measuring something"

Be real though, how often does that situation happen? Hard braking is much more likely to be caused by driver inattention and/or following too closely.

 

[ATPMSD]

I created an Excel spreadsheet to calculate the safety score. I cannot upload it, but I will walk you through the attached JPG. Please let me know if you believe there is an error. See post #3 above for the source of the values used.

Column A: The name of the event
Column B: The factor used by Tesla. Note that "Autopilot disengagement" is the biggest hit
Column C: The number of times the event occurred within the time period
Column D: This is Column B raised to the power of Column C. For example, first line is 1.014495^11=1.172
PCF is calculated by multiplying all the values in Column D together, and then multiplying by 0.68285
Finally, the score is 115.382324-22.526504*PCF

 

[Darmie]

I created an Excel spreadsheet to calculate the safety score. I cannot upload it, but I will walk you through the attached JPG. Please let me know if you believe there is an error. See post #3 above for the source of the values used.

Column A: The name of the event
Column B: The factor used by Tesla. Note that "Autopilot disengagement" is the biggest hit
Column C: The number of times the event occurred within the time period
Column D: This is Column B raised to the power of Column C. For example, first line is 1.014495^11=1.172
PCF is calculated by multiplying all the values in Column D together, and then multiplying by 0.68285
Finally, the score is 115.382324-22.526504*PCF

View attachment 714841

See if you can use a shared google drive location and then share the link. This looks good. I personally have yet to push the button. We lease they Y on Turo and well, you see where this may go.

 

[ATPMSD]

See if you can use a shared google drive location and then share the link

Tesla Safety Score B.xlsx

docs.google.com

 

[BCNY]

Be real though, how often does that situation happen? Hard braking is much more likely to be caused by driver inattention and/or following too closely.

My point is that unless that situation happens, I will have a perfect score. And if that situation happens and I do the right thing, I will be penalized.

 

[QS01]

My point is that unless that situation happens, I will have a perfect score. And if that situation happens and I do the right thing, I will be penalized.

The issue, for me at least, is that regen braking will cause a hard brake event! There is no way that the force of regen should be considered a hard brake.
Was going 30 mph and let off accelerator dropping me to 20 before approaching a school zone. That event cause a HB for me. Didn’t use the brake pedal the entire time.

their logic or force for a hard brake is not correct — and it should not be able to be flagged as such unless you actually hit the brake pedal.

 

[ATPMSD]

And if that situation happens and I do the right thing, I will be penalized

There is no such thing as the perfect metric. I do applaud Tesla for disclosing how the score is calculated.

The issue, for me at least, is that regen braking will cause a hard brake event!

If this is true, it seems like this feedback should be made to Tesla. Regen braking should defiantly not cause a hard brake event!

 

[Daniel in SD]

The issue, for me at least, is that regen braking will cause a hard brake event! There is no way that the force of regen should be considered a hard brake.
Was going 30 mph and let off accelerator dropping me to 20 before approaching a school zone. That event cause a HB for me. Didn’t use the brake pedal the entire time.

their logic or force for a hard brake is not correct — and it should not be able to be flagged as such unless you actually hit the brake pedal.

Did you use full regen on a steep downhill during your drive?

My point is that unless that situation happens, I will have a perfect score. And if that situation happens and I do the right thing, I will be penalized.

This is only an issue because you know the formula. Someone who drives in situations where they are constantly braking and swerving to avoid collisions is more likely to have a collision no matter how good a driver they are. The formula is actually "predicted collision frequency", it does not attempt to normalize for driving location.

 

[queue]

There is no such thing as the perfect metric. I do applaud Tesla for disclosing how the score is calculated.



If this is true, it seems like this feedback should be made to Tesla. Regen braking should defiantly not cause a hard brake event!

Same thing happened to me driving down a hill for approximately 1 mile, I was regenerative braking and was dinged significantly. WTF would have happened if I actually braked???

 

[MrsSmith]

I finally got my score (finally sideloaded the app). I was sure it would show a lot of hard braking, aggressive turning, and unsafe following. I was in a rush and decided to just drive like I normally would. Turning into sharp turns at normal speed (between 20-30mph) and using the brake to go into each turn. There are at least 3 sharp turns on the way to the destination and on the way back. For following distance, I don't ride people's bumpers but I was closer than 5 car lengths for a good portion.

The safety score noted the unsafe following distance (11.3%), Aggressive turning (1.5%) and 0 of anything else, with a total score of 98. So in my experience, it's a lot more forgiving than I anticipated from reading other people's experiences.

 

[Daniel in SD]

Same thing happened to me driving down a hill for approximately 1 mile, I was regenerative braking and was dinged significantly. WTF would have happened if I actually braked???

If you actually braked your score would probably have been better. It says it's based on time braking so braking hard for a shorter period of time is better than braking right over the 0.3g threshold for a longer period of time.
Better would be to not use full regen (keep very light pressure on throttle) on hills so you slow down but not quite as hard. I agree that this is tricky.

 

[Sporty]

Be real though, how often does that situation happen? Hard braking is much more likely to be caused by driver inattention and/or following too closely.

That’s not true. I just got 2 on a normal drive. One a pedestrian decided to walk across the crosswalk. They have the right away and you must stop. Second was on a 35mph road and a yellow light. It was morning & crowded and would have been unsafe to run. There’s no way that “hard stop” (which is not really all that hard) that could have been avoided unless you’re willing to drive below the speed limit.

 

[HighZ]

Does anyone have an idea how they came up with those factors? I get the Autopilot disengagement and hard breaking have the greatest impact, but my little brain has a hard time understanding why those numbers? I would have thought they could have come up with something simpler.

I would have also thought unsafe following would be higher, but maybe because it is not a consistent measure.

 

[Daniel in SD]

Does anyone have an idea how they came up with those factors? I get the Autopilot disengagement and hard breaking have the greatest impact, but my little brain has a hard time understanding why those numbers? I would have thought they could have come up with something simpler.

I would have also thought unsafe following would be higher, but maybe because it is not a consistent measure.

Unsafe following very often results in hard braking so its effect is already included in the hard braking metric. I'm not sure why they included it the formula at all since it's such a small factor.

The current formula was derived based on statistical modeling using 6 billion miles of fleet data. We expect to make changes to the formula in the future as we gain more customer and data insights.