Safety Score

[AlanSubie4Life]

Old habits die hard! 49.9% unsafe following, score of 99 for this 4-mile drive. Fortunately this will take about 3-4 seconds of following someone later today to patch up and bring my daily score up to 100 (at 33% unsafe following for the day right now), haha. Should take no more than about 5 minutes on the main road, if I get around to it. For this low mileage I'm not sure I'll bother. This was the only following event logged on the drive, of course.

It's 100% correct though. I like to stay out of the slow lane on this road because it's a bunch of manhole covers on the right hand side of the lane (you can't straddle them easily). That old habit of avoiding them (you can see I avoided one) really bit me. Should have waited!

Love the scoring system. It enforces excellent safety habits. Except for the AP incentivization as mentioned earlier.

Are the days weighted depending upon how many miles are driven in each day?

Yes, each day's score is first rounded to the nearest integer. Then the mileage-weighted average of the scores is calculated, and rounded to the nearest integer.

 

[brianman]

Based on a recent Elon tween, the beta firmware will not be released to beta button users until 10/8, not 10/1. So grannies, keep driving like old nannies for "two weeks" and keep those safety scores up! Apparently, Tesla needs a week to incorporate changes from the v10.1 EAP feedback, before releasing v10.2 to those lucky enough to have a perfect 100 score on 10/8.

That's my read as well. It's unfortunate that Elon tweeted clearly "a week needed" for eligibility and then played the "software is late" card. I frankly don't care if the software needs another week or another month, but moving the goal posts on eligibility is frustrating.

 

[brianman]

It was reported yesterday that Elon thinks the cutoff will be around 80%, but it could take forever to those in the 80's to see the actual download. Only a 1000 downloads per day, starting with those who have scored 100 and have a reasonable amount of miles accumulated. With 100's of thousands of beta button requests, many could be driving like nannies (or opting out) for many months.

That's odd that he wants to include "average" drivers in the program. Even worse when "scored average" (80%) is likely off from their model because of score awareness and driving behavior changes.

 

[phantasms]

It was reported yesterday that Elon thinks the cutoff will be around 80%, but it could take forever to those in the 80's to see the actual download. Only a 1000 downloads per day, starting with those who have scored 100 and have a reasonable amount of miles accumulated. With 100's of thousands of beta button requests, many could be driving like nannies (or opting out) for many months.

Where was this reported? Makes me feel a little better about my 94.

 

[brianman]

I read somewhere that for the FSD beta they were indeed using the internal camera to monitor driver attention and either ding them or disqualify them. It wasn’t speculation. I’ll see if I can dig that up.

For those actually using the beta that's plausible. The previous comments about were using that with vehicles/drivers as part of Safety Score. Doing so without explicit consent would be a really bad move on Tesla's part, especially when excluding it form the Safety Score page.

 

[Big Earl]

That's my read as well. It's unfortunate that Elon tweeted clearly "a week needed" for eligibility and then played the "software is late" card. I frankly don't care if the software needs another week or another month, but moving the goal posts on eligibility is frustrating.

He didn't guarantee access to FSD Beta after a week of good driving. He said that the past week of driving data needs to be good in order to qualify. When FSD Beta is ready for the next round of testers to be added, the past week of driving data will be used to determine eligibility.

If the release is scheduled for 10/8, driving data from 10/1 - 10/7 will be used to determine your eligibility.

 

[FalconFour]

Where was this reported? Makes me feel a little better about my 94.

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

 

[brianman]

> Or testing new firmware
don't have to drive slow/under the limit

> or gaming a score
don't have to drive slow/under the limit

> or recently got a ticket
even more to the point: don't have to drive slow/under the limit.

1. You don't know what you're testing. So you don't know either way.
2. You don't know what score they are gaming. So you don't know either way.
3. You don't know how stressed they are about the ticket and/or if they are overcompensating.

So -- without additional information -- your assertions are not viable.

 

[brianman]

He didn't guarantee access to FSD Beta after a week of good driving. He said that the past week of driving data needs to be good in order to qualify. When FSD Beta is ready for the next round of testers to be added, the past week of driving data will be used to determine eligibility.

If the release is scheduled for 10/8, driving data from 10/1 - 10/7 will be used to determine your eligibility.

Elon apparently never really guarantees anything. I wasn't talking about guarantees. I was talking about what he asserted. It's pretty clear:

"If driving behavior is good for 7 days, beta access will be granted."

The software availability date has nothing to do with this statement.

Apparently a newer tweet moved the goal post to be aligned with software availability which is what I find frustrating.

Tesla will open controversial FSD beta software to owners with a good driving record | TechCrunch

Tesla CEO Elon Musk said the company will use personal driving data to determine whether owners who have paid for its controversial “Full Self-Driving”

techcrunch.com

 

[Sporty]

He didn't guarantee access to FSD Beta after a week of good driving. He said that the past week of driving data needs to be good in order to qualify. When FSD Beta is ready for the next round of testers to be added, the past week of driving data will be used to determine eligibility.

If the release is scheduled for 10/8, driving data from 10/1 - 10/7 will be used to determine your eligibility.

I’m sure there’s going to be a thread where folks are auditing who’s getting in during the first round and who’s not. I wonder if we’ll see any out of score order accesses or not.

I personally hope so, but that’s mostly because I’d like to qualify. Don’t have a 100 but feel that my decisions on that were mostly good ones.

 

[diplomat33]

It occurred to me that with some modifications the safety score could be used to measure FSD Beta's safety. All the metrics, except for "forced AP disengagement", are good preventive metrics for safety. They would need to change the "forced AP disengagement" to "manual disengagements". But I could see Tesla using the Safety Score to gauge the safety of the system.

I also wonder if Tesla might start including the average safety score of users in their quarterly AP Safety reports.

 

[Sporty]

It occurred to me that with some modifications the safety score could be used to measure FSD Beta's safety. All the metrics, except for "forced AP disengagement", are good preventive metrics for safety. They would need to change the "forced AP disengagement" to "manual disengagements". But I could see Tesla using the Safety Score to gauge the safety of the system.

I also wonder if Tesla might start including the average safety score of users in their quarterly AP Safety reports.

context for some operations is missing. That would be useful to add for both cases. Eg hard-breaking is sometimes called for. Car cuts you off and responding correctly. Both should be +s in some contexts not -s.

I don’t know about aggressive turning. Overall I feel that its setting is too low as well, but maybe safely/probably dodging something.

 

[AlanSubie4Life]

Eg hard-breaking is sometimes called for. Car cuts you off and responding correctly. Both should be +s in some contexts not -s.

The argument would be that you should nearly never get cut off, because you should nearly always see that it is going to happen 3-4 seconds before it occurs. Obviously at the extreme limits there are things that are impossible to predict, but driving for the Safety Score over the past week, with extremely heightened vigilance, I'd say that over 95% of such events (which would have occurred before and would have required an abrupt reaction on my part) are predictable and I'm safely able to mitigate them with no sudden deceleration, etc.

My errors have been due to understanding the system thresholds initially, and errors since then have been due to inattention and carelessness - but this is a relative term - I would say I'm MUCH more aware of my environment now than I have been and I'm therefore much safer. The only issue is really whether I'm WILLING in those very rare unpredictable instances to hit the brakes, turn quickly, etc., or try to avoid doing so (this is unnatural and a negative byproduct of the scoring system).

As an example, a cat crossed my neighborhood street a few nights ago. Normally I would have been bombing along the street at 30mph. But since it was night and I know bunnies sometimes cross the road (I rarely see cats actually!) that I have to brake for, I had slowed to 20-25mph. As a result, I was looking for animals, saw the cat, was able to ease off the accelerator and slow down gently, and there was never a risk of killing it. So, slightly lower speeds in predictably risky situations, paired with heightened vigilance, led to a less abrupt stopping maneuver and was much safer for the cat.

As another example, yesterday evening I was in the #3 lane doing 75mph and saw in the rearview someone in a BMW coming up in the #2 lane doing about 95-100mph. I could see that they had no path through traffic ahead, other than to weave between me and the car in front, and transition to the #4 slow lane. So I eased off the accelerator substantially to open the gap (which was already 2.5 seconds) and make this maneuver easier for the driver as he approached. As he made the predicted transition, I engaged autopilot to mask the unsafe following (unnatural, but whatever). Overall, this was a safer event than it would have been otherwise, since I was vigilant and aware of what was happening around me, and traveling with a large gap to the vehicle in front.

 

[EVNow]

He didn't guarantee access to FSD Beta after a week of good driving. He said that the past week of driving data needs to be good in order to qualify. When FSD Beta is ready for the next round of testers to be added, the past week of driving data will be used to determine eligibility.

If the release is scheduled for 10/8, driving data from 10/1 - 10/7 will be used to determine your eligibility.

Definitely not what he said.

Most literal way to read the tweets is
- One gets a beta software request button.
- Once pushed, safety score will be assessed for 7 days
- The safety score is used to enroll 1,000 people per day starting with score of 100 and then going down
- The rollout starts 10/8 with the release of 10.2

Now - what actually happens, we don't know. It could be 7 days from the time button was pushed by an individual, some kind of cut off date before roll out etc.

 

[brianman]

The argument would be that you should nearly never get cut off, because you should nearly always see that it is going to happen 3-4 seconds before it occurs.

Speaking to this point...
In the last week I've been hyper-sensitive to cars in neighboring lanes and conga lines in adjacent lanes. Previously I was aware of them from a "be ready for evasive manuevers" but with the scoring it's pushed more toward "slow down even more when there's congestion". Table whether that's a good idea or not. It's some context.

When slowing down to 5 or 10 below the speed limit (not aggressively, but gradually), you'll naturally trigger more passing around you and some drivers (intentionally or not) aggressively return to your lane after moving out from behind you to an adjacent lane. I've found that those aggressive re-entries ("cutoffs") have not led to harsh scoring penalties.

(Note that none of the above is in the passing lane. I have strong opinions about proper use of the passing lane.)

 

[phantasms]

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

I saw that. More curious about this mention of “above 80”.

 

[brkaus]

And some idiot stopped at a stop sign and then pulled right in front of me on the main through street… resulting in collision warning, hard braking and aggressive cornering.

looks like a 2nd office for the idiot based on damage to front corner of their car.

 

[JosiahLuscher]

This driving/scoing app is open source and looks _very_ similar. I wonder if the Tesla team borrowed code from it:

Zenroad – Apps on Google Play

 

[TIppy]

I also wonder if Tesla might start including the average safety score of users in their quarterly AP Safety reports.

When I first starting using the app on android (side loaded immediately), there was a fleet average value for each of the five parameters. It was in a light gray that was hard to see. It was a little bar graph and a floating point average value. It was right under the bright green bar graph of each parameter. It's no longer there.

 

[JosiahLuscher]

I'm having trouble doing enough "good" braking to cushon moments of "hard" braking. I need immediate feedback. There are a bunch of accelerometer apps out there I'll be testing this one on the way home:

G Sensor - Apps on Google Play

It's like a G-ball running on Android

play.google.com

I've read a lot of the posts on this thread, but maybe I missed something. Any advice on how to impove braking numbers? I'm soo close to 99 I can taste it. If I can drive really well, I might be able to round up to 100 by the 8th.