Safety Score

[FalconFour]

See what I mean about AP being ridiculously laggy about acceleration/deceleration?? It wasn't always like that! Previous versions were dead-on about following distance and adapted quickly. But I imagine that responsiveness was at the cost of responding too quickly to "ghosts", false detections of merges, appearing cars, ... bridges, etc. So they smoothed the hell out of it, to an excessive degree, in knee-jerk response style. Now it follows "unsafe" close, though it knows it can respond quicker to true incidents. And it eventually regains an ... acceptable following distance, though I still hate how close it stays at "7", almost like the settings 2...6 are meaningless. 1 or 2 in the left lane where you should be passing, following, or moving over, and 5...7 in the other lanes. 7 in the right lane - when trying to leave room for mergers and exiters (or just taking it slow and not trying to shoo the following guy to go the speed limit) - should really be further away.

I just hope/suspect that Tesla is silently recording AP's mistakes and comparing it to human driving, and, well... this period of driving stats may be eye-opening to the AP team. If we get a 99 score driving ourselves, but AP can't even keep a cool 75... yeesh. To be a fly on the wall to the guy analyzing AP vs. human driving results... lol.

 

[FalconFour]

Ironically, in spite of all the complaints here, I think the thing that is the least safe about the Safety Score system is that it incentivizes AP use in busy situations, and the characteristically unnatural and terrible driving it produces. A true 100 score for me in my 40 miles of driving today though.

Probably one of the worst incidents I've ever had on AP. There's a reason I typically only use AP when there's light traffic! Yet with the Safety Score I must do the exact opposite. Sad.

As the description says (this was I-15S at the split to the 163):

If driving manually (which I would normally do in these circumstances; I was using AP to avoid unsafe following Safety Score dings) I would have signaled right and gotten to the right lane sooner, but the traffic in the fast lane was moving very slowly (it's always a disaster at this interchange, with people cutting in late, but normally traffic in the left lane does move the fastest, which is why I had picked it) and the lane to the right had a steady stream of cars - and Autopilot would have had a difficult time of it (but this is what I should have done, and just tried to have it find a spot - ironically I thought I would end up following too closely and wanted to wait). Going manual mode would have meant a Safety Score ding! The entire video is on Autopilot, with Lane Change Confirmation required, following distance 7. The only thing I did was reduce set speed (by about 10mph) immediately, as soon I saw the truck start to signal (I have no idea why he decided to get in so late - those familiar with this interchange know that you have several miles to get in the correct lane). And then Autopilot proceeded to ignore the truck and accelerate into it, after he was fully in my lane. I held my foot over the brake but did not use it (eventually AP applied the brakes/eased off accelerator); would have meant an unsafe following ding!

The wife was not happy - and neither was I. It would have been bad, but acceptable, if it had continued to open up space after the truck changed lanes (the correct behavior would have been for AP to see the truck signaling, and preemptively allow 2 seconds following when the truck entered the lane, which it's not capable of doing) . In retrospect I could have kept dialing back the set speed to 40-50mph, but it surprised me (I don't know why this was a surprise) that AP did not see the truck at all for a good long time. I expected it to continue to fall back, and its behavior was exacerbated by the truck also applying the brakes. Radar really doing a great job.

Perfect 100 safety score, 0.0% for everything. About 0.3 seconds following distance.

Preceding this, I had been in the right lane, but there was a lot of slow traffic which I had shifted left to avoid, and there were also disabled vehicles parked on the right shoulder which I wanted to avoid. Then the lead car started driving slowly, and then all the traffic started passing us on the right. In retrospect, should have just dealt with the traffic in the slow lane.

I blame the Prius.

Tesla should have skipped the part of the Safety Score where AP masks errors.

Oof... I'm just gonna grace-quote this post and bring it some visibility on the next page before it gets lost. A lot of work went into this

 

[Lowchrg]

Evaluation period confirmed pushed to 10/8 with Beta 10.2 release.

One more week of driving like grandma.

https://twitter.com/x/status/1443892118745911309

 

[nvx1977]

Oof... I'm just gonna grace-quote this post and bring it some visibility on the next page before it gets lost. A lot of work went into this

you know you've been on the forums too long when you know spot 20 on each page is the worst.

 

[brkaus]

View attachment 716597 Chick fill a drive thru caused the 21.1%

Strange. I did Chick-fil-A with a long line and had a perfect score that drive.

 

[dduffey]

Good morning TMC'ers. Goal is to get to 100 overall today, and I need 85 miles @ 100 daily score.

Wife just told me to take kids to school (two different schools), requires travel on busy roads > 50 mpg this morning during rush hour AND it's raining (no autopilot). Wish me luck!

Day	Score	Miles
Saturday	93​	9​
Sunday
Monday	96​	21​
Tuesday	99​	4​
Wednesday	100​	107​
Thursday	100​	76​
Friday
	Avg. Score	Total
	99.30415​	217​

 

[dduffey]

Good morning TMC'ers. Goal is to get to 100 overall today, and I need 85 miles @ 100 daily score.

Wife just told me to take kids to school (two different schools), requires travel on busy roads > 50 mpg this morning during rush hour AND it's raining (no autopilot). Wish me luck!

Day	Score	Miles
Saturday	93​	9​
Sunday
Monday	96​	21​
Tuesday	99​	4​
Wednesday	100​	107​
Thursday	100​	76​
Friday
	Avg. Score	Total
	99.30415​	217​

Edit: I chicken-littled out. It's dark and raining, half the cars don't have their lights on, it's a 55mph road so everyone is nascar'ing at 65+. Several folks cutting in just to cut back out. But what really got me was a squirrel that decided to try and run across the road which I had to brake for. Even the squirrels are idiots today. I'm taking that as a sign to stay home

 

[jebinc]

Strange. I did Chick-fil-A with a long line and had a perfect score that drive.

Vision car vs radar car, I assume.

 

[Skunis]

Evaluation period confirmed pushed to 10/8 with Beta 10.2 release.

One more week of driving like grandma.

https://twitter.com/x/status/1443892118745911309

Well, this stinks! I do not always drive my car working from home so have been making time to just go out and drive like 5 miles a day. So now we have to drive for 2 weeks or will they just use 7 days and make a decision a week later?

Tesla did finally roll out the official 4.1 Android update. Got it this morning and my score is 99!

 

[mltv]

Vision car vs radar car, I assume.

My radar does ding a number of times against the curb. All other days my drives are all lower single digit (< 5). One thing is certain, I am picking up next time instead of going thru the drive thru. It was a good 20 minutes stuck in the drive thru.

 

[mongo]

Yellow light? Better to run it than to brake.

Especially if by braking you get rear ended by the car behind you. Or the road surface is poor and you won't actually stop in time.

Sharp curve? Better to cut across the yellow line to make it less sharp than take the "aggressive turning" hit.

Or take the curve at a lower speed.

Get cut off? Better to brake gently and nearly rear-end than to brake hard at a safe distance.

Or be more alert for bad drivers and increase your buffer so you don't need to brake hard.

 

[Kingoglow]

Seven days with a 100 straight.
Unsafe Follow is sitting at 9% but still shows as green.

Looks like I need to keep it up for another seven days though.

 

[boonedocks]

View attachment 716597 Chick fill a drive thru caused the 21.1%

guess you wouldn't consider that a "blessed" day

 

[kbM3]

As I said previously, going up a hill with a 30 percent grade will cause a deceleration greater than 0.3 gs. But he is going down hill, so the acceleration due to gravity is try to increase his speed, not decrease it. That is, it is accelerating him. So by properly applying the brakes, he can descend the hill while decelerating less than 0.3 g.

There will be a constant amount of braking to just cancel the gravitational force, and an additional braking force to slow the car down. It is this additional application of brakes that needs to be applied smoothly.

I assume they aren't just using the accelerometers along the long axis of the car. Their dead reckoning algorithms should be taking into account the actual change in velocity along the car, the size of the resultant acceleration vector and the rotational accelerometers.

Going downhill causes the accelerometer to measure deceleration, not uphill. Think about it. If you’re at a stop on a steep decline, gravity is pulling you forward against your seatbelt, exactly like braking would on level ground. On a steep incline, gravity is pushing you back in the seat just like acceleration.

 

[kbM3]

I had a 97 (98? Don’t quite remember) going into yesterday. Yesterday I did a 600 mile road trip to a vendor. Two stops to charge and the stop at the vendor. And a couple used-coffee disposal breaks at rest stops. The stretch to the first rest stop had 0 hard brakes and that included all the stops I made getting out of Chicago. Then I had to brake for a yellow light, and I guess I braked “hard” to get into the turn lane for a turn. I don’t know exactly, but none were dangerous. I didn’t care too much cuz I thought those were my only few “hard” braking events I’d ever logged. Well, now I have a 94 cuz yesterday’s 93 is mileage weighted against all the other days.
This is just stupid statistics. You can’t average averages - that’s a statistics cardinal sin. It should be hard braking events per total braking events or hard braking events per mile driven, but the fact I drove 600 miles yesterday and probably had fewer braking events than I do in a normal 60 mile day shouldn’t make yesterday’s braking events more consequential just cuz I drove more that day. I call bullshit

As explained by many here, It IS hard braking time / overall braking time. The problem is time on AP doesn’t count. Time on AP acts like an amplifier in the sense that it weights either a good score, or a bad score more heavily in your overall.

E.g. 1 s of hard braking with no other braking, followed by a 500 mile AP Trip would give you a horrible score that would wreck your week.

conversely 1 mile of perfect driving followed by a 500 mile AP trip would give you 501 miles worth of a 100 score for that day.

 

[boonedocks]

I would really like to see Tesla's fleet wide A/P score. If I am relegated to bounce between 98-99 and seeing how A/P reacts but doesn't count against me.....I would bet fleet wide A/P scores would be in the low-mid 80's

 

[kbM3]

I h


I agree, but we are all speculating right now. I was just trying to understand why some events are masked and others are not for some people. I had a FCW while on AP on a country road (caused by shadows across the road), but it was not counted.

Regarding the camera, the opt-in button did say that the camera would be used to detect attention. I have already gotten the "pay attention" alert if I am looking away for too long, maybe 10 seconds. And Tesla has never said what "using AP appropriately" even means. If they don't explain it, it already seems like a 'sly' remark. Do you know what it means (you can speculate)?

If you have ever looked away from the road for 10 seconds, not only should you NEVER get in the FSD Beta, you should have AP removed from your vehicle.

 

[helvio]

If you have ever looked away from the road for 10 seconds, not only should you NEVER get in the FSD Beta, you should have AP removed from your vehicle.

Whoa, take a chill pill. Perhaps the guy was cruising through the desert where nothing ever happens. Which would be perfectly fine. You never know.

 

[scottf200]

See what I mean about AP being ridiculously laggy about acceleration/deceleration?? It wasn't always like that! Previous versions were dead-on about following distance and adapted quickly. But I imagine that responsiveness was at the cost of responding too quickly to "ghosts", false detections of merges, appearing cars, ... bridges, etc. So they smoothed the hell out of it, to an excessive degree, in knee-jerk response style. ...

It needed to be more elastic and smooth as it was unnaturally jerky and made passengers car sick. This was frequent complaint by passengers who then didn't want the driver to use AP at all.

 

[dduffey]

Weather improved and made it to 100 score (from 93 last Saturday). Another glamour shot.