Safety Score

[AlanSubie4Life]

Don’t know if they just average the daily scores to get your weekly safety score but I’d guess yes

" We combine your daily Safety Scores (up to 30 days) into a mileage-weighted average to calculate the aggregated Safety Score, which is displayed on the main ‘Safety Score’ screen of the Tesla app."

So I interpret this to mean they are not simply going to look at all the data on say, unsafe following, and figure out that overall percentage, and incorporate that (and the same for the other elements) into the formula.

This makes for some interesting optimization strategies. For example, let's say you had some very bad hard braking, 5% or something, on your first drive of the day. Subsequently, you would not just want to do a lot of miles on that day on AP at a consistent speed - you'd also want to make sure you got a lot of good braking (presumably you could not use AP for this braking improvement process, since those events are masked, whether good or bad, when AP is "used appropriately") on that day to bring down that % number for that day. Otherwise you'd be in a situation where just a very small number of events gets inflated by all those AP miles (with no braking) when it comes to the mileage-weighted Safety Score . If it's a very bad result, and you really don't think you'd be able to get the number down even with some more driving, you might well want to avoid driving for the rest of that day, to avoid having the weight of those mistakes be high in the aggregated safety score. Whatever you do, you have to make sure that you bring down those percentages, and the more miles you've got on that day, the more important that those % numbers are low.

If they were just doing the aggregate of each event type over the entire window the result could be very different. In that case, the only thing AP miles would do is help with overall masking of events, and also bring down the FCW number.

 

[Sporty]

The definition says you can only get that when moving at least 50 MPH.

View attachment 714598

Are you going 50+ MPH in the drive-thru line?

Other side of town .. all non-highway, multiple stop lights. The entire drive was below 50mph.

 

[EVNow]

They may already be cleansing the data. I also suspect that a ML model would be much better but then it would be a black box. Of course that might not be a bad thing since people changing their driving to score well on the model negates the stated purpose of the model.

Yes, they may be cleansing data - but apparently not enough.

BTW, why would ML given model be a black box ? The software basically gives you the best parameters to use and the coefficients.

 

[brkaus]

Anyone know if the 7 consecutive days of good driving scores is rolling? If on day 3 you hit a run-away shopping cart at Publix does your Safety Score keep averaging until you finally meet the required score or is it "game over"?

As long as you crush the cart but do not slam on the brakes or do a quick evasive turn…. You’re good!

 

[sperkin]

To help with aggressive turning, make sure to use all lanes and drift out wide on exit. Use the full road even though it’s illegal, but it gives a better score.

I took a left turn on a 50mph down hill road and turned into lane 1. It was very smooth so was surprised to see a ding. I think I could have done better if I hit the apex and turned into lane 3. I couldn’t brake harder without triggering a hard brake, but I didn’t take that turn very fast either. I rode the brake all the way into the turn and even on exit to keep it slow and smooth. I,m just gonna turn wide and use all 3 lanes next time.

 

[kbM3]

Looks like we get a score for each day of driving. So if you get a low score early in the day you can go out and drive more error-free miles and the score for the day increases. Don’t know if they just average the daily scores to get your weekly safety score but I’d guess yes. So you might be able to salvage a bad daily score by adding more miles that day.

Fairly sure, it would be calculated based on total number of infractions / total miles. Just averaging daily scores would give you wrong numbers.

Days are weighted by how many miles, so it is a weighted average.

 

[kbM3]

" We combine your daily Safety Scores (up to 30 days) into a mileage-weighted average to calculate the aggregated Safety Score, which is displayed on the main ‘Safety Score’ screen of the Tesla app."

So I interpret this to mean they are not simply going to look at all the data on say, unsafe following, and figure out that overall percentage, and incorporate that (and the same for the other elements) into the formula.

This makes for some interesting optimization strategies. For example, let's say you had some very bad hard braking, 5% or something, on your first drive of the day. Subsequently, you would not just want to do a lot of miles on that day on AP at a consistent speed - you'd also want to make sure you got a lot of good braking (presumably you could not use AP for this braking improvement process, since those events are masked, whether good or bad, when AP is "used appropriately") on that day to bring down that % number for that day. Otherwise you'd be in a situation where just a very small number of events gets inflated by all those AP miles (with no braking) when it comes to the mileage-weighted Safety Score . If it's a very bad result, and you really don't think you'd be able to get the number down even with some more driving, you might well want to avoid driving for the rest of that day, to avoid having the weight of those mistakes be high in the aggregated safety score. Whatever you do, you have to make sure that you bring down those percentages, and the more miles you've got on that day, the more important that those % numbers are low.

If they were just doing the aggregate of each event type over the entire window the result could be very different. In that case, the only thing AP miles would do is help with overall masking of events, and also bring down the FCW number.

I posted this a few hours ago.

 

[EVNow]

The definition says you can only get that when moving at least 50 MPH.

View attachment 714598

Are you going 50+ MPH in the drive-thru line?

There was another report of unsafe following because of queuing at McDonalds. Ofcourse could have been just an assumption ...

 

[EVNow]

Days are weighted by how many miles, so it is a weighted average.

Weighted avg would yield the same result as total infractions / total miles.

Ps : For metrics that use fraction of time, they should use total time for proper calculation.

 

[AlanSubie4Life]

I posted this a few hours ago.

I missed that, but yes, I just went back and saw your post and I agree. Seems like it was worth reiterating in any case.

Personally I think it's kind of a weird way to do it, but whatever. In practice if people weren't trying to game the system and were driving the same way all the time, it probably wouldn't make much difference which way they calculated the overall score. But since people are gaming it, it definitely matters.

 

[Sporty]

Anyone know if the 7 consecutive days of good driving scores is rolling? If on day 3 you hit a run-away shopping cart at Publix does your Safety Score keep averaging until you finally meet the required score or is it "game over"?

I don’t think it will penalize you for hitting a shopping cart - unless you swerve or hit the brakes. So just drive right through it…

 

[kbM3]

Weighted avg would yield the same result as total infractions / total miles.

For FCW or AP Jail that’s true, but not for the ratios like hard braking.

 

[EVNow]

Why do I have this feeling, someone is going to get into an accident because they are trying to get a good safety score ?

 

[AlanSubie4Life]

Weighted avg would yield the same result as total infractions / total miles.

Only for FCW I think. Not sure about AP jail; that is just 0 or 1 so it is not a rate (but maybe I'm not thinking about it right).

EDIT: It is different for AP jail.

For AP forced disengagements (which of course should never happen so this is not really important), you just have a score of 1. So once you get one, you just have to stop driving immediately unless your other scores are truly awful. More AP forced disengagements can't make your score for that day any worse (95 or lower is what you'll have which is awful of course), but you can make the weight of that day worse. So I guess the right strategy is if the other factors are making your score way worse than 95, it might make sense to drive a short distance to make them a lot better, but generally additional miles are a terrible idea (since 95 is an awful score).

This is different than the 1/numdays exponent I was saying yesterday (before I really understood the mileage-weighted average part of the calculation). Now it makes sense that in the app you can only specify 1 or 0. There's no other valid exponent. That goes into the Safety Score, and then from there it's mileage-weighted into your overall score.

 

[EVNow]

For FCW or AP Jail that’s true, but not for the ratios like hard braking.

Yes, instead of total miles, they should use total time for those.

Weighted avg by only miles for those metrics is weird calculation.

 

[kbM3]

Is it weird that I went for a long drive today early before traffic, to get a lot of miles at a 100 safety score (to make a combined 97 with my 90 score yesterday), and I’m taking an Uber to supper so I won’t screw it up?

 

[AlanSubie4Life]

Is it weird that I went for a long drive today to get a lot of miles at a 100 safety score (to make up for the 90 yesterday), and I’m taking an Uber to supper so I won’t screw it up?

Yes.

 

[BCNY]

This scoring sucks. So if there is a bad event out of your control in front of you (VERY often here in Miami, unfortunately) you will get penalized for being a safer driver and braking hard or swerving to avoid an accident...

In the meantime, the guy in the Tesla crossing 5 lanes, doing an illegal turn, crossing double yellow, etc doesn't get penalized.

Very stupid scoring system.

All times where my collision warning went off, or I had to brake really hard, was because something completely unexpected happened and braking hard or swerving was the safest thing to do.

For what it matters, I started driving in 1990 and only had 1 accident in my life, due to a guy under the influence crossing a red light for him and causing the accident.

I also ride motorcycles, with zero accidents on record. Never even fell once.

When you Lear how to ride a motorcycle, one of the things you are thought is how swerving can be life saving.

Yet with this idiotic scoring system, you will get penalized for avoiding an accident.

I enrolled for the beta, I will not change my driving, and I just need to hope Florida drivers don't mess up in front of me.

 

[Rob Washington]

After some experimenting, I suggest not to use autopilot except for straight freeway driving. I got dinged hard for aggressive driving, hard braking, and close following all on autopilot during a drive on back roads

 

[Daniel in SD]

Yes, they may be cleansing data - but apparently not enough.

BTW, why would ML given model be a black box ? The software basically gives you the best parameters to use and the coefficients.

I must admit that I know very little about ML. Are you saying that ML software could spit out a human readable formula to calculate predicted collision frequency?