Safety Score

[nvx1977]

Which is the generation that didn't grow up with video games (boomers?) ...

hah. yeah I guess I should have defined that better. Anyone who played Atari (or programmed one) in their formative years counts as "someone who grew up playing video games." Could be most of us!

Boomers most definitely did not grow up playing video games. Starts with the GenXers, of which I am a part of.

 

[FloridaJohn]

They need vast amounts of high quality data and this is what beta testers provide. Tesla has a team of 1,000 people doing manual labeling for FSD. I imagine this team has been able to keep pace with the 2,000 beta testers. But Tesla has been working on and improving automatic labeling. This is an essential part of their path forward. When machines handle half the labeling, Tesla can double the beta pool. When machines handle 90% of the labeling then the pool can increase by a factor of ten. And so on.

Tesla is planning on adding 1000 new testers to the beta pool every day. They will double the existing pool in two days and will be ten times larger in just 20 days. I haven't seen anything that leads me to believe Tesla has automated the labeling to that extent.

I think Tesla will do what they are now with the "fleet" data. They send out requirements to the cars to look for certain situations, and have those cars upload the clips when they come across that. I feel that it is unlikely Tesla is in a position to handle incoming data for tens of thousands of beta testers. This is most likely a PR move more than anything else.

 

[BitJam]

What's different about the stream of data from the beta testers vs. the stream of data from a few thousand carefully selected cars from the fleet? If the difference is that that stream of data from the beta testers will have information on where interventions took place, and bug reports from users tied to video clips...how is that not just a good way to find out what FSD bugs (limitations due to labeling, path planning, etc.) to focus work on (and work on labeling in those specific situations, etc.)?

As I explained, one big difference is the beta testers are running the latest FSD software. I also explained why streams of video data from the entire fleet are not collected and not useful. I tried to explain Tesla's need to vastly increase the number of beta testers. The beta fleet not only tags problem areas it also collects data from those problem areas instead of just tagging them. It's a brilliant way of collecting exactly the data needed instead of just identifying problem areas and then somehow magically sifting through millions of times more data (which is not and cannot be collected).

How is this different from what you originally said? It's different because it explains why they need to drastically increase the number of beta testers.

 

[EVNow]

hah. yeah I guess I should have defined that better. Anyone who played Atari (or programmed one) in their formative years counts as "someone who grew up playing video games." Could be most of us!

Fun fact : Never heard of video games growing up in India - (I'm part of GenX too).

Now even 2 year olds play vodeo games ...

ps : Though I should say - being competitive is not unique to gamers. Its been part of DNA for a billion years.

 

[BitJam]

Tesla is planning on adding 1000 new testers to the beta pool every day. They will double the existing pool in two days and will be ten times larger in just 20 days. I haven't seen anything that leads me to believe Tesla has automated the labeling to that extent.

I think Tesla will do what they are now with the "fleet" data. They send out requirements to the cars to look for certain situations, and have those cars upload the clips when they come across that. I feel that it is unlikely Tesla is in a position to handle incoming data for tens of thousands of beta testers. This is most likely a PR move more than anything else.

I agree with you that adding 1,000 new testers every day *indefinitely* is probably not going to happen. OTOH increasing the size of the beta team is essential for Tesla's FSD plans, as is automated data labeling. Using the Safety Score is a brilliant way to quickly increase the number of testers as safely as possible. I cannot believe this is just an elaborate hoax to gin up free publicity. It's a necessary step on their path to true FSD. Tesla is not Nikola.

 

[HighZ]

being competitive is not unique to gamers.

Very true. But I don't really care if I'm first, I just want to be in the first 1000. Not too much to ask, especially on my birthday this weekend. Never programmed an Atari, but did have some fun back in the day with my TRS-80 and Apple II.

 

[nvx1977]

but did have some fun back in the day with my TRS-80 and Apple II

same era. it counts CLOAD, CSAVE to the tape player!

 

[AlanSubie4Life]

They need the FSD beta software in those cars to see how FSD would actually react and what the person thinks is the correct action.

Understood. That sounds like finding bugs. Which may require labeling, or it may require improvements to the path planning, etc.

How is this different from what you originally said?

That wasn't me. I'm not sure, not going to speak for him, but the way I interpreted that post was that he was describing something that is basically identical to what you have outlined is happening (which is why I asked the question of you - it seems like you're both saying something very similar!). Specifically: having FSD beta testers allows Tesla to clearly identify bugs...

That post I think was in reference to needing "vast amounts of data" - and that post said that Tesla already has the capability of doing this - which is true (hence their campaigns when they want to look for something specific). And specifically, then it was stated that "FSD Beta testers just show them what bugs to work on." (This would not require a vast amount of data. And to me it seems like exactly what you're saying they're doing! Latest FSD software, specific clips, specific data from vehicle, in specific situations, etc. )

It does seem like they have plenty of bugs to work on already, which is what @Daniel in SD appeared to be saying - you are saying they no longer have enough bugs and need to expand the pool, to more diverse environments, or whatever they need. I guess that's where the difference in opinion lies, if I'm understanding correctly.

In any case I hope they expand the pool to the people with good Safety Scores, on the schedule they have indicated, whether they actually need to or not.

 

[MarcG]

Does anyone know if drives taken while a car is in service mode count toward the Safety Score? I’m taking my car in soon and they will likely drive it to test a couple of issues I have while driving.

 

[jebinc]

Does anyone know if drives taken while a car is in service mode count toward the Safety Score? I’m taking my car in soon and they will likely drive it to test a couple of issues I have while driving.

I believe I got charged with hard braking while getting my tires changed at the SC in service mode. I recall thinking that and almost posted a warning about it.

 

[AlanSubie4Life]

Does anyone know if drives taken while a car is in service mode count toward the Safety Score? I’m taking my car in soon and they will likely drive it to test a couple of issues I have while driving.

Good luck having them put it in service mode though! I assume they do as a matter of routine - but no guarantees in life.

 

[boonedocks]

Any of you here by chance figured out how to actually brake our cars?

Regen is so strong in these new cars (no settings for changing anything either) that just fully releasing the go pedal will ding you for a “Hard Braking”.

There is effectively no way to actually apply brakes to offset the dings by actually braking.

I am even tempted to put my car in Neutral on the expressway just to be able to actually hold the brake pedal for braking.

Never touch the brakes at all and still have .4 “Hard Braking” with 1,238 miles driven. (All of which are daily driving / commuting….no trips)

 

[AlanSubie4Life]

that just fully releasing the go pedal

Don't do that, especially not on a significant downhill!

still have .4 “Hard Braking”

If you're sure you did not touch the brakes at all, that probably happened on a hill.

I haven't touched the brake (except when already stationary) in several days.

Perhaps the new Model S is even stronger regen. In a way this is good - but it will require a lighter touch.

To offset the dings you'll have to do something other than pulse width modulation of the accelerator. You definitely do not need to apply the brakes to offset the dings.

 

[boonedocks]

Don't do that, especially not on a significant downhill!


If you're sure you did not touch the brakes at all, that probably happened on a hill.

I haven't touched the brake (except when already stationary) in several days.

Perhaps the new Model S is even stronger regen. In a way this is good - but it will require a lighter touch.

To offset the dings you'll have to do something other than pulse width modulation of the accelerator.

Oh I haven’t fully released the go pedal since the first day of driving and seeing how much it affected the HB score. 1,200+ miles in and still stuck at .4 for that one

 

[EVNow]

I'm trying to figure out whether I can go down the hill to "low lands". 500" in 1 mile.

ps : That seems to be about 10% gradient.

Success ! I went down the big hill to the low lands - and didn't get dinged. I drove slow (~30 mph on a 35 mph max road) - and saw the yellow traffic light at the bottom early and gently slowed down - still thought might get dinged.

BTW, one thing I realized when going up the hill I shouldn't release the accelerator quickly. That would make the natural "g" get added to regen which could get you over the 0.3g limit. Low on regen seems to be about 0.2 g. Braking while going downhill is actually not that bad - since the gravity is acting in the other direction making the deceleration slower.

Don't do that, especially not on a significant downhill!

Actually uphill ? Releasing the accelerator would cause significant deceleration uphill, rather than downhill - like I explained above.

 

[mhan00]

Did another 70 miles the last couple of days, a lot of it in traffic in downtown, on freeways, or in LAX, and I’m honestly starting to wonder how people drive to get nailed all the time. I got a 99 yesterday and 100 today. The 99 was because when I was driving home at night, I admittedly zoned out a little while driving on a section of road with no other cars and no cross streets and had to swerve around a large road kill that was illuminated at the last second. I got dinged for a hard turn event. Today got a 100 with no issues despite another day dealing with a ton of traffic while driving my aunt to LAX. I had originally thought my high score was because of my area, but It just doesn’t seem that hard to maintain a score in the high 90s even when driving in urban environments and busy freeways.

 

[AlanSubie4Life]

That would make the natural "g" get added to regen which could get you over the 0.3g limit.

Actually uphill ? Releasing the accelerator would cause significant deceleration uphill, rather than downhill - like I explained above

No that's not how it works. The gravity vector is pulling back on the car (you're being pushed BACK in your seat, as though you are accelerating) when going uphill. That's helpful, because the Safety Score is looking for a feeling of being pulled forward in your seat!

So uphill is easier on the limits. No worries about releasing the accelerator suddenly when going uphill. It's like the opposite of downhill!

If you want to verify, try a different car (might need to be manual transmission or EV I guess) with a G meter and do the same, on an uphill. Then compare to downhill.

It all makes sense since it's easier to stop going uphill, so more rapid (wheel speed) stops should be allowed. And they are as far as I can tell.

Obviously it's still possible to trigger a violation on an uphill. But you'll have to apply the brakes.

 

[MarcG]

Braking while going downhill is actually not that bad - since the gravity is acting in the other direction making the deceleration slower.

Uhm.. what?? I think you got your physics backwards.

Edit: yeah. Alan beat me to it.

 

[nvx1977]

I have done full regen on an uphill and did not get dinged. I noticed my speed slowing down very rapidly, so from a pure deceleration perspective, it probably exceeds what 0.3g on flat ground would do. But the sensor is measuring g-forces, not raw deceleration. This stuff is hard to wrap your head around if you've never thought about it, especially the g force vectors!

 

[azred]

I wonder what is the higher risk in AZ. FSD Beta or all the anti-vaxxers / anti-maskers

Probably the “anti’s“ by a not narrow margin. My wife is a Southwest flight attendant based in PHX and she flew this week with a rabid “anti” flight attendant. The “anti” was raging about other airlines and their mandatory vax rules. And then the next day after her rant, Southwest announced that they too were mandating vax! Ouch.