Safety Score

[gearchruncher]

In other words, it may be the area(s) you are driving aren't fit for FSD testing at this point.

Yep. Yet Tesla lets FSD beta to many users in very dense areas, and acts like the Safety Score is a measure of YOU as a driver, not your location. I mean some of the most popular videos include it going down Lombard street and driving in downtown Seattle (and turning right into a group of pedestrians).

I mean, this is actually CITY STREETS AUTOSTEER, right? And the place you shouldn't be using it is in cities?

 

[momo3605]

Wouldn't "hard braking a couple times a day to prevent hitting a car jumping in front of you or braking to turn into a parking lot without indicating at the last minute, or for jaywalkers" mean that you are at an increased risk to a collision from someone that is driving and doesn't have to hard brake a couple times a day to avoid a collision?

In other words, it may be the area(s) you are driving aren't fit for FSD testing at this point.

I commented on this above, but Tesla has given Gali, TeslaRaj, and others FSD beta in city centers with heavy traffic, and that was earlier versions which were presumably not mature, why do those testers get to have the FSD beta?

But i do generally agree with the premise that testing FSD in lower traffic scenarios is safer, sure.

 

[gearchruncher]

How could it possible have anywhere near that amount of collision warnings for such a short, slow, properly spaced drive with no audible alerts.

It's per 1000 miles. So 101.9 per 1000 miles.

If you drove 3 miles, and had one FCW, there you are. Unless you have FCW on "Early" or "medium" you can get hits for alerts that you never hear (and they can all be false positives)

Admittedly the math doesn't work. One FCW in 3 miles would be one every 333 miles out of 1000. There's some serious issues with trying to look at this in any kind of short trip, where single random events can be incorrectly assumed to repeat.

It's really not a good idea for Tesla to share data like this with people that aren't really up on the statistics. Anything like this really only works over thousands of miles and many, many days. 7 days is way too short, much less one drive, especially against the stakes of getting a $10,000 software package you paid for years ago without this condition present at the time of purchase.

 

[nvx1977]

I really question how it’s registering hits and then how it factors them into the calcs.

I’ve had the button three days and have something like 7 trips so far.

All of my days and trips have been 98 and 99 so far.

Then I went to check today and it says I’m NINE today, bring my 3 day average form 98/99 down to 48 !!!

I checked the specifics for each trip and there’s one thing driving it. For one 3 mile trip out of 3 trips for the day…

System states 101.9 on Forward Collision Warnings.

It’s a 3 mile trip and I received no audible warnings. There was nearly no hard breaking and ZERO Unsafe Following.

How could it possible have anywhere near that amount of collision warnings for such a short, slow, properly spaced drive with no audible alerts.

As someone else explained, it's extrapolating based on 1000mi. You can mitigate the hit by logging more miles.

also fyi: hard braking and hard cornering cannot be mitigated by logging more miles, as the calculation is limited to periods of deceleration and periods of lateral g's.

 

[momo3605]

As someone else explained, it's extrapolating based on 1000mi. You can mitigate the hit by logging more miles.

also fyi: hard braking and hard cornering cannot be mitigated by logging more miles, as the calculation is limited to periods of deceleration and periods of lateral g's.

so in order to mitigate hard braking, you can drive around at night accelerate and use regen to slow down and keep cycling for a few miles?

 

[jeremymc7]

Of all my trips expect one my score has been 98/99.

One of my trips today had Forward Collision of 101.9 was a slow, gentle, properly space drive with no audible alerts. That trip scored at NINE and took my 99 average down to 48 immediately.

NO IDEA had this is even possible. Seems like a system problem somehow.

 

[momo3605]

Of all my trips expect one my score has been 98/99.

One of my trips today had Forward Collision of 101.9 was a slow, gentle, properly space drive with no audible alerts. That trip scored at NINE and took my 99 average down to 48 immediately.

NO IDEA had this is even possible. Seems like a system problem somehow.

someone already answered your question above, the 101.9 is an extrapolation over 1000 miles. Meaning if you only drove 10 miles and had 1 FCW, that would be something like a score of 100 (out of 1000) miles.

 

[22522]

Selection will likely include some driving competence indicator.

 

[pshifrin]

Of all my trips expect one my score has been 98/99.

One of my trips today had Forward Collision of 101.9 was a slow, gentle, properly space drive with no audible alerts. That trip scored at NINE and took my 99 average down to 48 immediately.

NO IDEA had this is even possible. Seems like a system problem somehow.

Ok this is very strange. I was driving home today like Grandma. Not a single warning to be had, on AP most of it and I also got dinged with a 101.9. Simply not possible! Same, all drives since I got this were 97-100.

 

[jeremymc7]

someone already answered your question above, the 101.9 is an extrapolation over 1000 miles. Meaning if you only drove 10 miles and had 1 FCW, that would be something like a score of 100 (out of 1000) miles.

Not sure of that. I I’m looking at the daily trips details for each day. The 101.9 is only on one trip at 3 miles. Trips (and days) before and after are all fine. So it’s got a problem with that one trip. It also doesn’t explain how a slow trip with no alerts have 101.9, especially when all other trips and days had zero alerts.

 

[AlanSubie4Life]

The fleet wide average in the Safety Score app was 91 as of yesterday.

Yes, but to be clear, the actual median Beta Queue scores are likely much higher, as these numbers are probably from the entire fleet before the release of The Button. They should add a line in the plots for the median Beta Queue performance, plotted over time.

There are many times when you need to brake hard or swerve quickly around kid, pedestrians, bikes, or anything that suddenly showed up on the street.

In addition to the location considerations mentioned already…I would add:

Admittedly, this does happen, rarely, to even the most careful and expert drivers. But there is something wrong with defensive driving techniques if this is something that happens many times in a week. It should be something like a 10% chance it happens in a given week (the current interval of evaluation), probably, for a good driver.
I’ve found with the way I drive on the Beta Queue that this would almost never happen now - because I’m constantly scanning the road, side roads, parked cars, etc., up to 1/4 of a mile ahead for hazards which I may need to respond to. This is a good thing! I even give parked cars a huge wide berth on curves where possible, to avoid any chance of FCW (probably also much safer).
Generally speaking, things do not just “show up” on the street. It CAN happen - especially at night with animals. But maybe Tesla wants to avoid releasing to people who drive a lot at night in rural areas. These are less safe situations and would generally show up as such in the Safety Score.

System states 101.9 on Forward Collision Warnings.

This 101.9 value has been showing up in a number of screen captures from people so I actually think it is some sort of bug. In addition, for the people who have mentioned it, for the posted trip distances, there’s no way to get the number of FCWs to work out to be even close to an integer. It is possible that the display cap is 101.9 and your actual value is ~333 though.

If you have FCW set to Early and you had no warning in the vehicle at all, the actual event itself may also be spurious.

 

[jeremymc7]

If you drove 3 miles, and had one FCW, there you are. Unless you have FCW on "Early" or "medium" you can get hits for alerts that you never hear (and they can all be false positives)

I always have it on early. Is it better to have it on late, at least as far as the scoring? Or does it only matter for the warning and not for the scoring.

 

[AlanSubie4Life]

I always have it on early. Is it better to have it on late, at least as far as the scoring? Or does it only matter for the warning and not for the scoring.

Only matters for the warning; it uses medium for scoring. Use early to give yourself a chance to avoid dings. (Would only help at the margins, and will likely result in hard braking, so it’s a tradeoff.)

 

[jeremymc7]

Taking this conversation in another direction what are the thoughts of having AutoPilot on.

When using AP I notice more harsh breaking, or using actual breaks, vs my own driving which I try and get every drop of regen.

Do you get dinged when AP does something. Or are things on AP ignored. Are miles with AP on counted for things like hard breaking or ignored.

I’ve been using AP less while in the 7 day period. But I question if not using AP at all during the 7-day period is a good or bad thing in terms of scoring.

 

[jeremymc7]

[QUOTE="gearchruncher, post: 5960842, member: 52712”Can you imagine the outrage if speeding counted against you? They clearly are avoiding using some metrics for marketing reasons.
[/QUOTE]

Speeding could be used if it was accurate. I often find the posted speed is inaccurate relative to what the car thinks it is. This has also been a common complain with the people that have already been running the current beta’s.

 

[AlanSubie4Life]

Do you get dinged when AP does something. Or are things on AP ignored. Are miles with AP on counted for things like hard breaking or ignored.

Supposedly they are ignored and they likely are on the freeway. There is an open question about whether use of AP on surface streets has the same masking effect (there is a caveat about “appropriate use”). Everyone is dying for a well controlled experiment to find out.

So AP on the freeway is probably a good idea especially in heavy traffic where cut-ins can reduce your score (unsafe following). But AP has other limitations and if you’re forced to brake because AP is being dumb about seeing things, that will hurt you (a lot - hard braking is very bad). You also have to be careful about entry and exit from AP disrupting your driving “flow state.”

I am using AP selectively, and I have successfully used it to mask cut-ins. I drive manually a lot of the time, but engage AP as soon as I think I am going to have someone cut-in. And other times where it is low risk of AP misbehavior or disrupting my flow state, I just use AP (only on the freeway though!), because I figure Tesla wants to see that.

 

[gearchruncher]

Generally speaking, things do not just “show up” on the street.

Except for those darn yellow lights, particularly where that driver in front of you decides they need to stop when you would have gone though (because braking is bad, mmmkay).

 

[gearchruncher]

Speeding could be used if it was accurate. I often find the posted speed is inaccurate relative to what the car thinks it is.

Forward Collision Warning could be used if it was accurate. I often find it goes off for false positives, and lots of others have reported this as well.
...Oh wait. They are using it despite many false positives. Weird.

Makes you wonder how accurate their estimate of follow distance is, given they are using that too, and on modern cars all they have is vision, which is "solved problem" except for the fact that they still won't let you drive as fast with vision only cars as radar cars.

This is just one big lottery with extra steps.

 

[AlanSubie4Life]

Except for those darn yellow lights, particularly where that driver in front of you decides they need to stop when you would have gone though (because braking is bad, mmmkay).

Yes, which is where my only major braking demerit showed up so far. This can be mitigated by driving slowly (speed limit or below) and maintaining proper distance (3 seconds) from vehicles in front. You can also look at side streets to predict whether the light is likely to change and either accelerate or slow down as appropriate.

Most lights are timed such that if you are traveling the posted speed limit or slower, there is plenty of time to amble to a stop.

Yellow lights are my biggest concern and stress with my Beta Queue driving. They are the hardest thing to anticipate, but there are ways to predict them (looking 1/2 mile ahead where possible helps too).

This is why it would be nice to know if AP masking on surface streets works, since you could just let AP come screeching to a halt for you. But it is so punitive to experiment (you have to be below 0.2% to maintain a score of 100) that someone with a bad score has to do the careful short experiment.

 

[qdeathstar]

I’m surprise they don’t check for seatbelt use. Seems like something that should count.