Safety Score

[MarcG]

I just looked into how Tesla transmits data to the app via API for the aggregate Safety Score page, and I found some very interesting information that may not have been shared here before.

I'm including a snippet of the data params below, but here are some takeaways I gathered:

• There's possibly a "drive rating" metric coming, though it's currently showing 0
• Besides miles driven, Tesla is potentially counting hours driven, though it's currently showing "null"
• The "raw" numbers for each metric are transmitted with lots of decimals via API, but truncated to 1 decimal in the app
• Tesla is also able to report which decile for each metric I'm sitting in (top 20% for most, 10% for braking for me)
• However the xxxDisplayDecile params are all "null" for now, so perhaps they will eventually show up in the app
• "Aggressive turning" is referred to as "harsh cornering" in the API, which is interesting choice of words
• Tesla is likely already tracking longitudinal acceleration variance, but it's currently showing up as "null"
• There are also metrics for ABS events including one triggered by user (likely the driver), also both "null" for now

Here is an easy-to-read version of the API response I see for my account (grouping by metrics is mine):

"driveRating":0
"safetyScore":99
"milesDriven":550
"driveHours":null

"autoSteerStrikeoutsCount":0
"autoSteerStrikeoutsDecile":0
"autoSteerStrikeoutsRaw":0

"forwardCollisionWarningsDecile":0.2
"forwardCollisionWarningsCount":null
"forwardCollisionWarningsAvg":1.8
"forwardCollisionWarningsMedian":10.2
"forwardCollisionWarningsRaw":1.81891

"headwayRatioAvg":2.5
"headwayRatioDecile":0.2
"headwayRatioDisplayDecile":null
"headwayRatioMedian":15.8
"headwayRatioRaw":2.46292

"harshBrakingPercentAvg":0.2
"harshBrakingPercentDecile":0.1
"harshBrakingPercentDisplayDecile":null
"harshBrakingPercentMedian":1.8
"harshBrakingPercentRaw":0.206897

"harshCorneringPercentAvg":0.8
"harshCorneringPercentDecile":0.2
"harshCorneringPercentDisplayDecile":null
"harshCorneringPercentMedian":2.8
"harshCorneringPercentRaw":0.837775

"longitudinalAccelVariance":null
"absEvents":null
"userAbsEvents":null

 

[AlanSubie4Life]

So it is better to drive more on the same day to increase your score for that day if you had a FCW, not stop driving?

Complicated rule of thumb I guess would be fewer than 3-4 miles, stop driving immediately (the FCW score capping is helping you a lot!). Otherwise, drive as much as you can to get your score as high as you can, definitely more than 40-50 miles, but pay attention to how many miles EXACTLY that you drive. Due to the quantization (rounding), you only want to barely reach the next integer score, and not go beyond the required miles (because each additional mile beyond that step will really hurt you in your quest for 100, unless you get to the NEXT score).

So for example, drive the exact distance you need to to get to 97 (96.5), then immediately stop, if that's all you have time for. If you have time, drive to 98 (97.5), but then immediately stop (so important to stop). Examples of how much this can help in your case (the 46 miles vs. 48 miles example) are given above. The score simulator is your friend (but using a non-rounding Excel formula is even better).

Hopefully @yurmix is still following along! I guess this is a good summary.

 

[gearchruncher]

Fleet familiarity is what will make FSD safer than humans

So you're saying the majority of human accidents are caused by the inability to see the permanent road ahead or signs, and lack of familiarity with the local road, not distraction, speeding, etc?

Hot take.

The real world isn't exactly like Inception where roads just morph randomly on a daily basis.

You live in CA, yet don't deal with construction and variable speed limits all the time? How can FSD deal with a construction zone, but then be incompetent at reading standard signs on the highway?

 

[macmee]

I just looked into how Tesla transmits data to the app via API for the aggregate Safety Score page, and I found some very interesting information that may not have been shared here before.

I'm including a snippet of the data params below, but here are some takeaways I gathered:

• There's a potential "drive rating" metric coming, though it's currently showing 0
• Besides miles driven, Tesla is potentially counting hours driven, though it's currently showing "null"
• The "raw" numbers for each metric are transmitted with lots of decimals via API, but truncated to 1 decimal in the app
• Tesla is also able to report which decile for each metric I'm sitting in (top 20% for most, 10% for braking for me)
• However the xxxDisplayDecile params are all "null" for now, so perhaps they will eventually show up in the app
• "Aggressive turning" is referred to as "harsh cornering" in the API, which is interesting choice of words
• Tesla is likely already tracking longitudinal acceleration variance, but it's currently showing up as "null"
• There's also a metric for ABS events including one triggers by user, also both "null" for now

Here is an easy-to-read version of the API response I see for my account (grouping by metrics is mine):

"driveRating":0
"safetyScore":99
"milesDriven":550
"driveHours":null

"autoSteerStrikeoutsCount":0
"autoSteerStrikeoutsDecile":0
"autoSteerStrikeoutsRaw":0

"forwardCollisionWarningsDecile":0.2
"forwardCollisionWarningsCount":null
"forwardCollisionWarningsAvg":1.8
"forwardCollisionWarningsMedian":10.2
"forwardCollisionWarningsRaw":1.81891

"headwayRatioAvg":2.5
"headwayRatioDecile":0.2
"headwayRatioDisplayDecile":null
"headwayRatioMedian":15.8
"headwayRatioRaw":2.46292

"harshBrakingPercentAvg":0.2
"harshBrakingPercentDecile":0.1
"harshBrakingPercentDisplayDecile":null
"harshBrakingPercentMedian":1.8
"harshBrakingPercentRaw":0.206897

"harshCorneringPercentAvg":0.8
"harshCorneringPercentDecile":0.2
"harshCorneringPercentDisplayDecile":null
"harshCorneringPercentMedian":2.8
"harshCorneringPercentRaw":0.837775

"longitudinalAccelVariance":null
"absEvents":null
"userAbsEvents":null

Is there an endpoint you're hitting to get this? I have a little tesla apple watch app I've been working on in swiftUI. If you know the endpoint I'd love to hit it on my own account to pull down my own metrics

 

[EVNow]

Finished my drive for today - unfortunately got dinged for hard braking. Overall still 100 - but 99 for today. Probably happened downhill where I got a yellow light and decided a tad late to stop instead of turn or on a pedestrian walking where someone pushed the button when I was close and had to stop (or both !). Wasn't expecting to be dinged - otherwise would have done the 2 finger solute. Both times I thought I had come to a slow and gradual stop.

 

[powertoold]

Regarding phantom FCWs, I wouldn't be surprised if Tesla's software could figure out if phantom vs not (for purposes of safety score beta), like "only count FCW if the lead car has been visible for 2 seconds." This would eliminate all the phantom FCWs that last 0.2 seconds or whatever.

 

[gearchruncher]

Regarding phantom FCWs, I wouldn't be surprised if Tesla's software could figure out if phantom vs not (for purposes of safety score beta), like "only count FCW if the lead car has been visible for 2 seconds." This would eliminate all the phantom FCWs that last 0.2 seconds or whatever.

Yeah, this is what the company that is supposedly about to release FULL SELF DRIVING CAPABILITY on the world should be working on. Filtering out false positive FCW's from a stupid made up score that they are using to restrict access to FULL SELF DRIVING, not you know, actually fixing their detection methods which are needed to actually do FULL SELF DRIVING.

This whole score distraction is an amazing, mind blowing distraction technique made up by someone with a deep understanding of human psychology, and it's working better than I bet they even guessed it would.

 

[TIppy]

When braking on a downgrade, normal regen will give you a hard braking ding

Are you using your brakes or just regen? EVNow says when he is using just regen on a hill it takes him longer to stop from a given speed than when he is on level ground. Do you find this to be true? That would indicate that Tesla is just using an accelerometer aligned with the direction of travel, but would also mean that you shouldn't get dinged while just using regen on a hill. It would mean, however, that the gravitational acceleration is added to your braking acceleration.

This would complicate stopping in time on a hill. If autopilot knows how far away the stopped car ahead of you is and your current speed, it can calculate what g force is require to stop in time. It would then use a combination of regen and brakes to generate that amount of deceleration. If it's using an accelerometer as feedback for how hard to apply the brakes on a hill, it won't stop in time.

 

[Sporty]

People who are wondering why speeding isn't included? Well the NNs aren't reliable enough to read the signs, so it's difficult to be certain someone is speeding or not. Also, hard breaking, aggressive turning, and FCW will usually be worse if someone routinely speeds.

If you’re passing by all the other driving cars that the NN identifies….

 

[_jal_]

Yeah, this is what the company that is supposedly about to release FULL SELF DRIVING CAPABILITY on the world should be working on. Filtering out false positive FCW's from a stupid made up score that they are using to restrict access to FULL SELF DRIVING, not you know, actually fixing their detection methods which are needed to actually do FULL SELF DRIVING.

This whole score distraction is an amazing, mind blowing distraction technique made up by someone with a deep understanding of human psychology, and it's working better than I bet they even guessed it would.

Yes! It’s his own personal version of The Real World. It’s kind of weird that thousands or maybe tens of thousands of people have literally turned their daily commutes into a video game to win a prize

 

[_jal_]

Got it, thanks for the explanation. So to answer poor @yurmix 's question that started this whole conversion: it IS better to drive more on the same day (assuming no more FCWs) to increase your score for that day if you already had a FCW, and NOT stop driving for the day, right?

I'm curious why Tesla chose to use rounding that way instead of true numbers.. maybe it's to make it easier/more fair for someone not analyzing and dissecting numbers like we are

But, in my case, I couldn’t possibly recover. I did a 600 mile round trip to a vendor. I made four stops along the way each with maybe a dozen braking events. I’m getting off the highway and going to Culver’s to charge or whatever. But so I’m rolling at 20 MPH in a strange place with no one around me and I see the obstructed driveway that I’m about to pass so I put on the brakes! Or I’m on a rural highway with a super short left turn lane that I need to decelerate quickly in without causing a traffic fiasco! Well, due to nothing other than “hard braking” events that I couldn’t give a crap less about I’m at a 93 for the day. But I’d driven 600 miles that day. Now each day cuz I live in a city I probably broke 10 times that much even though I drive 1/10 as many miles. Don’t matter! I got a 93 despite hours of AP on the highway so that one score weighs as much as a two weeks worth of workdays. The math is poor

 

[thewishmaster]

Yes! It’s his own personal version of The Real World. It’s kind of weird that thousands or maybe tens of thousands of people have literally turned their daily commutes into a video game to win a prize

Prediction: the final selection criteria will annoy everyone equally, both higher and lower scores. Feels like there’s no way they’ll stack-rank by score and go down the list…

 

[MarcG]

Feels like there’s no way they’ll stack-rank by score and go down the list…

Why wouldn’t they? It’s essentially what Elon wrote..

 

[gearchruncher]

Why wouldn’t they? It’s essentially what Elon wrote..

And a few days before that, Elon wrote that EVERYONE with a score above a single threshold WILL get the beta.
And it was going to be 10.1, last weekend.
Now it's some sort of 1K people per day. 10.2. Next weekend.
Tomorrow it will be????

 

[gearchruncher]

to win a prize

A prize that you paid for, and was sold:
1) Not as a lottery
2) As available years ago

 

[thewishmaster]

Why wouldn’t they? It’s essentially what Elon wrote..

He writes many things

 

[K2500Z71]

Still at 100 and probably have driven 400+ miles since the safety score started. Not particularly surprising since I use AP/auto steer whenever I can and submit bug reports whenever it has an issue. This is probably what Elon wants us to do when we get the beta as it will help accelerate the development.

 

[gearchruncher]

I can and submit bug reports whenever it has an issue.

Ok, who's going to tell him what happens to bug reports?

This is probably what Elon wants us to do when we get the beta as it will help accelerate the development.

It's not the way that current beta works, but I'm sure there will be clear, concise written instructions on how to use FSD beta, and how to report issues, and then Tesla will give good release notes as they fix and change things, just like they always do.
Right?

 

[jebinc]

Ok, who's going to tell him what happens to bug reports?


It's not the way that current beta works, but I'm sure there will be clear, concise written instructions on how to use FSD beta, and how to report issues, and then Tesla will give good release notes as they fix and change things, just like they always do.
Right?

Yeah, I was going to, but didn’t want to burst their bubble of glee!

 

[EVNow]

Prediction: the final selection criteria will annoy everyone equally, both higher and lower scores. Feels like there’s no way they’ll stack-rank by score and go down the list…

Why ?

Looks like wishful thinking to me.