Safety Score

[kbM3]

I'm getting constantly dinged for "hard braking" when ever I leave my neighborhood, which is down a very steep hill. The deceleration on the downslope is triggering false hard braking events. Even if I slow down extremely gradually and drawn out, I'm fighting against gravity and Tesla's g sensor.

My safety score daily trips show consistent hard braking for all trips leaving my neighborhood, and none for trips returning.

Anyone have a pointer for where I can send this feedback to Tesla? It seems to be a major gap in their measurement algorithm. They need to take into account the angle of approach when braking.

If possible, engage AP on that road.

If not, and no one is behind you, then braking harder is better. Hard braking is counted in seconds spent over 0.3 g’s, not in how much over 0.3g’s you go.

Or go down that road in reverse, which also eliminates forward collision warnings as a bonus.

 

[DaveG_NJ]

It would be intesting to see the breakdown of scores by state. TeslaFi shows it by car model, which isn't that interesting (to me at least). I'd like to see the numbers for NJ where "safe following distance" is defined in inches. Cars routinely speed into your nice safe space then slam on the brakes to time their next move (usually going from the passing lane, through the center lane to the right lane about 3 feet before their exit).

And AP is not the answer, it will blithely speed at 80 mph next to cars that are panic braking in an adjacent lane. I've seen one too many Tesla crash videos posted by Green where someone cut off a Tesla and the car can't react in time, because, well, it's not a human who can spot trouble literally a mile away. Sometimes I feel that FSD and human drivers cannot co-exist.

 

[THXULTRA]

Going to be interesting to see how long several days is for us with a 99 score...

 

[jebinc]

Going to be interesting to see how long several days is for us with a 99 score...

"days"?

 

[EVNow]

It would be intesting to see the breakdown of scores by state.

In the spreadsheet there are 80+ 100 scorers from virtually every state.

What would be interesting is to see the stats by state of people who are not tracking the safety score (there by modifying the behavior). But I guess the biggest difference would be between city / sub-urban / ex-urban driving in all the states.

 

[KArnold]

can you please clarify if a Model S originally built with AP2.0 (from late 2016 or ealry 2017) and later upgraded (FSD HW3 + MCU2) is eligible to be part of the easrly access FSD Beta?
I am not able to find evidence of such configuration with a safety score of 100 that received 2021.32.25

Thank you in advance.

Tweeted essentially the same to Elon.. I'm not expecting a response from any of these sources.

 

[Ruffles]

Going to be interesting to see how long several days is for us with a 99 score...

There are also people with 100 still waiting. Hopefully we all get in the next batch. I wish we could get an update on how Tesla thinks its going so far. From what I've seen on Twitter and in other places, the new testers confirm that it certainly is beta still but they are also pleasantly surprised. I've yet to hear anyone scream that FSD tried to kill them and Tesla needs to shut it down (that will come from Jalopnik of course ). I would love to see Tesla proceed with adding additional testers day by day.

 

[Wlagrange]

Anyone find that when you enter your data into the score calculator that the percentage doesn’t exactly correspond? I have 0.1% hard braking, 0.7% aggressive turning and 2.9% close following with no forward collision warnings or ap disengagements.

The calculator says 100, but I have 99 on the app.

 

[THXULTRA]

There are also people with 100 still waiting. Hopefully we all get in the next batch. I wish we could get an update on how Tesla thinks its going so far. From what I've seen on Twitter and in other places, the new testers confirm that it certainly is beta still but they are also pleasantly surprised. I've yet to hear anyone scream that FSD tried to kill them and Tesla needs to shut it down (that will come from Jalopnik of course ). I would love to see Tesla proceed with adding additional testers day by day.

That is all I wish is that Tesla would be more up front about timing. How long is several days? At least say we expect to push out to the next group of 100 scores today and the 99 next week or something. I still have my complaints about the whole safety score thing too but that is another conversation....

 

[tivoboy]

Going to be interesting to see how long several days is for us with a 99 score...

I think we might have to look to Einsteins laws of relativity to answer this.. on Mars, days and weeks could be months and years ON EARTH.

(okay, not really but I’m never taking EM word for “daily” to not mean something like - in SEVERAL MONTHS, one DAY we’ll roll it out to people with 99, then a few MONTHS later, on ANOTHER DAY, we’ll roll it out o 98’s.)

it’ll probably be faster than that, but I’m just holding out for surprise rather than, well, actually holding out.

 

[EVNow]

That is all I wish is that Tesla would be more up front about timing. How long is several days? At least say we expect to push out to the next group of 100 scores today and the 99 next week or something. I still have my complaints about the whole safety score thing too but that is another conversation....

They have not decided. They are going to decide based on feedback from the 1,000 who got the beta yesterday.

If they could already decide and announce, there would be little need to change the 1,000/day they were talking about earlier.

 

[helvio]

Anyone find that when you enter your data into the score calculator that the percentage doesn’t exactly correspond? I have 0.1% hard braking, 0.7% aggressive turning and 2.9% close following with no forward collision warnings or ap disengagements.

The calculator says 100, but I have 99 on the app.

The calculator is for daily trips, your score is for multiple days

 

[n.one.one]

Anyone find that when you enter your data into the score calculator that the percentage doesn’t exactly correspond? I have 0.1% hard braking, 0.7% aggressive turning and 2.9% close following with no forward collision warnings or ap disengagements.

The calculator says 100, but I have 99 on the app.

Give some consideration to putting your daily scores and miles in my spreadsheet.

Tesla Safety Score - Tracking & Planning v10

Safety Score This is set to read only. To make your own sheet go to the file menu, 'Make a Copy' and open in Google Sheets (not in Excel). No share requests please. There are hidden columns with some of the working formulas so be careful what you change or move. <a href="https://teslamotorsclu...

docs.google.com

 

[THXULTRA]

They have not decided. They are going to decide based on feedback from the 1,000 who got the beta yesterday.

If they could already decide and announce, there would be little need to change the 1,000/day they were talking about earlier.

Goes back to my whole complaint about them not planning the rollout. First the safety score was only for 7 days. Then they were rolling it to 1k per day... Now it is we will see how it goes and several days later... AT least give some sort of daily update... Does anyone even know the safety score needed to stay in the beta once you get it?

 

[EVNow]

Goes back to my whole complaint about them not planning the rollout.

You seem to think this whole thing can be done in some kind of "waterfall" method with tight planning. You are wrong. Software industry has found out a long time back that "waterfall" just doesn't work well - since its not possible to accurately forecast time needed for software features. That is just for normal business applications - let alone bleeding edge AI software using CNN that has basically not been ever done before.

The only people who will "win" this race are the people who are very agile.

ps : All the "plans" even Musk (or any number of other CEOs/Execs) talked about in the media have all come to nothing. But you still think "plans" are useful.

 

[tivoboy]

You seem to think this whole thing can be done in some kind of "waterfall" method with tight planning. You are wrong. Software industry has found out a long time back that "waterfall" just doesn't work well - since its not possible to accurately forecast time needed for software features. That is just for normal business applications - let alone bleeding edge AI software using CNN that has basically not been ever done before.

The only people who will "win" this race are the people who are very agile.

ps : All the "plans" even Musk (or any number of other CEOs/Execs) talked about in the media have all come to nothing. But you still think "plans" are useful.

Good thing they manage it with an Agile model.

 

[THXULTRA]

You seem to think this whole thing can be done in some kind of "waterfall" method with tight planning. You are wrong. Software industry has found out a long time back that "waterfall" just doesn't work well - since its not possible to accurately forecast time needed for software features. That is just for normal business applications - let alone bleeding edge AI software using CNN that has basically not been ever done before.

The only people who will "win" this race are the people who are very agile.

ps : All the "plans" even Musk (or any number of other CEOs/Execs) talked about in the media have all come to nothing. But you still think "plans" are useful.

I know my fair share about software testing... I also know that Tesla has been testing FSD for some time and has a really good idea of how well it works in the real world. They could very easily say if all goes well we will push to the next group this Friday. They had some set backs and waited to do this push on Sunday but at least let people know. I'm a Tesla fanboy and I can even admit communication is one area Tesla needs to work on.

 

[techlogik]

Looking back should have just gamed the system and got a 100 instead of actually trying to drive my car as a daily commute in a major metro area....stuck at 98, about impossible to get it higher now. Too bad I figured out the 2 button reset hosed the tracking on the first day and didn't just go with it. Oh well...

 

[EVNow]

I know my fair share about software testing... I also know that Tesla has been testing FSD for some time and has a really good idea of how well it works in the real world. They could very easily say if all goes well we will push to the next group this Friday. They had some set backs and waited to do this push on Sunday but at least let people know. I'm a Tesla fanboy and I can even admit communication is one area Tesla needs to work on.

Absolutely they can work on communications.

But do you think - "if all goes well we will push to the next group this Friday" is useful ? It just sets up expectations that if not met, will create more disappointment.

We have to decide whether we want
- target dates that gets missed often (nature of fsd software at this time)
- not announce dates until its clear when it can happen

I see people beat up on Tesla for doing both of those.

BTW, I don't think the below remark is not correct. They really don't know how the software works with a large number of non-employees - esp. how well people can monitor. Its not just about how the software works - but also about the human-machine interface.

I also know that Tesla has been testing FSD for some time and has a really good idea of how well it works in the real world.

 

[THXULTRA]

Absolutely they can work on communications.

But do you think - "if all goes well we will push to the next group this Friday" is useful ? It just sets up expectations that if not met, will create more disappointment.

We have to decide whether we want
- target dates that gets missed often (nature of fsd software at this time)
- not announce dates until its clear when it can happen

I see people beat up on Tesla for doing both of those.

BTW, I don't think the below remark is not correct. They really don't know how the software works with a large number of non-employees - esp. how well people can monitor. Its not just about how the software works - but also about the human-machine interface.

I do think a if all goes well we will push to the next group on Friday is much better then saying several days. If they have issues they just communicate we are having issues with the current rollout and have to delay rolling out to more people. We will continue to give updates expect a update Monday or whenever...

Whole point of the safety score (as flawed as it is) is to test that the human's they choose will monitor the system. Also people that don't monitor the system can be easily booted from the beta. Finally if they have that level of concern about having a public beta then they shouldn't be having one at all.