Blog Musk Gives Details on FSD 10.2 Beta Release

Tesla will release the next version of it’s Full Self Driving Beta on October 8, Chief Executive Elon Musk said in a tweet.

The company released last week version 10.1 of the software, which included a button to request access to the beta. Tesla previously only gave access to employees and select owners.

By requesting access, owners give Tesla permission to evaluate their driving to create a Safety Score. The Safety Score is an assessment of driving behavior based on five metrics called Safety Factors. These are combined to estimate the likelihood that driving could result in a future collision.

https://twitter.com/x/status/1442716005936078855

Musk said drivers will need to have a near perfect score in the early roll out.

“First few days probably 100/100, then 99, 98, etc.,” Musk tweeted.

https://twitter.com/x/status/1442707528924684293

Tesla’s self-driving system is currently under scrutiny from federal investigators after a series of crashes when Autopilot was engaged.

 

[fiveten]

While on AP for most of a 50mi commute this morning I have 60%+ Unsafe Follow Distance (still rated the trip at a 99 score lol). All non-AP driving was at less than 50MPH which supposedly will never trigger USF. I don't think they know WTH they're talking about. Theres some disconnect between the engineers that wrote the thing and the tech doc publishers that wrote the description page

Did you use the accelerator at any point when you were on autopilot? I've seen Musk say when using autopilot "correctly" you won't get dinged, but perhaps if you use the accelerator and get too close while on autopilot it will ding you? Or maybe when using when not on a divided highway it could ding you?

 

[techlogik]

While on AP for most of a 50mi commute this morning I have 60%+ Unsafe Follow Distance (still rated the trip at a 99 score lol). All non-AP driving was at less than 50MPH which supposedly will never trigger USF. I don't think they know WTH they're talking about. Theres some disconnect between the engineers that wrote the thing and the tech doc publishers that wrote the description page

The current AP system clearly has no clue, especially with vision only, how to accurately discern distance. Everybody is getting dinged on following too closely. I have it on 7, drive like a granny, and when hand driving, I'm sooooo far back from people, even while on AP, it should never ding me.

EXCEPT!!! When stopped, the car launches from 0-15mph like a banshee and is on the arse of the bumper in front of you, then it just lifts completely off the accelerator, and starts going at a snails pace...that is probably part of the following too closely.

Next, when slowing/stopping, the car can never get that correct and will jam on the brakes way too late, or if rolling slowly in stop go, it is hit/miss how far it will stay away from the car ahead of you. I'm constantly doing what the car should be able to do, I'm slowing it by scrolling the wheel back every few seconds as I see traffic stopping/slowing ahead making ACC slow down and keep it's distance and not slam on the brakes like it will do. Then I ramp the wheel back up, or just stop it from AP while stopped to make it quit launching stupidly. People are getting so furious with the cars behavior behind me. It launches, then it crawls, people dive bomb in front of you. The car is doing things that make it more dangerous as a driving on the road for you and others. Anybody that denies this must be worried about their stock price in TSLA. No half-sensible person would start/stop like the car currently is doing. Hence, the follow too close is starting to add up on me.

I seriously have no idea how I can follow any further and do any better using AP "correctly" since all my hits on following too closely have to be due entirely to AP/FSD itself.

With that said, I'm still hanging onto a 100 right now...going to start hand driving more in particular situations, especially when stopped, so it won't launch and ride the bumper in front of me or drive too close even on 7 in particular situations. Otherwise, it is out of my control as usual. The car is calculating something in particular situations, while rolling slowly or who knows what...to think you are an unsafe driver.

 

[kono]

I didn’t pay for fsd with the conditions that Must imposes and changes week to week. Obviously his delay pattern is turning the table on the owners.

 

[kono]

I did not purchase fed with the conditions that Musk has been imposing and changing week to week. He has obviously acted in now turning the table with the burden on the Owners for more delay while behaving like Trump in tweeting and lambasting Ms. Homendy (NTSB). I see a class action suit forthcoming.

 

[Two-rocks]

What difference does it make how you brake? A G is a G.

It called ‘covering the brake’ and considered and advanced driving skill apparently. Also requires human intuition to use, since you are covering the brake in anticipation of something yet to materialize.

 

[buyleonard]

The current AP system clearly has no clue, especially with vision only, how to accurately discern distance.

I'm seeing the opposite. If I drive on AP, even on city streets, I get a score of a 100. And my car slows and stops smoothly, with plenty of distance to the next car. No idea why it would be so different, maybe re-calibrate your cameras?

 

[_jal_]

Or drive badly or by misfortune get dinged - but make up for it by driving on freeways with AP on for a few hundred miles.

I've posted in a cpl places already, but even this strategy is flawed. I drove 600 miles yesterday but had a couple of "hard braking" dings. Total BS, but whatever cuz I drove so many miles. But the proportion of braking events yesterday was higher than normal (I usually drive around the town and have hundreds I assume of braking events.) Yesterday I had few total events cuz it was only when getting off the highway and driving to the destination. So my score was 93! Total BS. Anyhow, cuz that day had so many miles compared to my usually daily 60 or so miles, it is weighted 10 times as much as the other days in the aggregate!

In statistics, you can't average averages. That's like rule #1, but that's what they're doing. They should be aggregating all the braking events across all of the days and using the rate of "hard braking" events against that. My hard braking events of yesterday shouldn't count more just cuz I drove hundreds of miles between them

 

[EVNow]

I've posted in a cpl places already, but even this strategy is flawed.

I saw your other posts.

Since some of the parameters are % of time calculations - but daily ratings are mileage weighted for total rating - it gets very weird.

Yes - going on long trips to improve the score is risky.

 

[_jal_]

I saw your other posts.

Since some of the parameters are % of time calculations - but daily ratings are mileage weighted for total rating - it gets very weird.

Yes - going on long trips to improve the score is risky.

Between my credit score, my Uber passenger rating, my kids’ opinion of my cooking, and my TMC reaction score, I have plenty of people judging me. I’m just going to drive from now on

 

[HighZ]

That is exactly what I am referring to as the "flaw". I'd say that using regen / one pedal driving is an indication of a driver's (well above average) safety consciousness and should be rewarded, not penalized.

I agree - I think that the safety score will go through a few 'improvements' as I am sure it will be with us for some time, this being one of them. I think they need to increase the G force necessary to create an event. 3g is a bit light. Maybe it is 3.5g or something else. But I expect some changes will occur as they get all of this feedback.

 

[EVNow]

Between my credit score, my Uber passenger rating, my kids’ opinion of my cooking, and my TMC reaction score, I have plenty of people judging me. I’m just going to drive from now on

I agree - Beta Testing is not for everyone.

I'm also surprised, like Musk, that so many people are interested in beta testing. Thats because people think this is "beta" like the highway NOA. Its not - this is really early access beta. People need to be a lot more careful with this FSD than probably even their usual driving.

 

[Dan D.]

I agree - Beta Testing is not for everyone.

I'm also surprised, like Musk, that so many people are interested in beta testing. Thats because people think this is "beta" like the highway NOA. Its not - this is really early access beta. People need to be a lot more careful with this FSD than probably even their usual driving.

1) There's likely a higher number of IT/engineer/geeks that drive Tesla's - and I mean that in a good way - who understand product testing. That is exciting to them
2) Self-driving is a pretty cool thing. If it was a Beta test for wiper activation frequency tables I doubt you'd get a lot of takers
3) People paid a lot of money and want to get the thing they paid for - even if they don't understand how to be a beta tester
4) Some will test it safely, some don't know how to test but will see what happens, some will abuse it, some will activate it without any idea what is going on. The owner will not be the only one driving the car
5) Are those on TMC's forum not representative of the Tesla public at large? Maybe most Tesla drivers typically don't get all that excited about things (I tend to think this can't possibly be true)?
6) I'm actually surprised that Musk is surprised. Is he really misreading his customers that much?
7) I'm still very glass-is-half-empty that this release is going to happen to ~1000 a day for many days, starting ~Oct 8th, hope something doesn't screw up the release

 

[GlmnAlyAirCar]

It called ‘covering the brake’ and considered and advanced driving skill apparently. Also requires human intuition to use, since you are covering the brake in anticipation of something yet to materialize.

This is too funny to warrant a response.

 

[EVNow]

5) Are those on TMC's forum not representative of the Tesla public at large? Maybe most Tesla drivers typically don't get all that excited about things (I tend to think this can't possibly be true)?

TMC users are definitely the more engaged and interested of the users. So, don't think they are typical at all.

BTW, even among TMC users autopilot/FSD is not a very high priority as you can see from forum usage.

 

[rsreeni@proton]

I have clicked the request button, but my app is NOT showing the safety score, even after uninstalling and installing multiple times. I clicked the button on Sunday, its been 5 days and still no show in the app.. anyone happen to have a similar problem ? - Sreeni

 

[GlmnAlyAirCar]

I have clicked the request button, but my app is NOT showing the safety score, even after uninstalling and installing multiple times. I clicked the button on Sunday, its been 5 days and still no show in the app.. anyone happen to have a similar problem ? - Sreeni

It's addressed on many threads. But what worked for me was force closing and re-opening the app.

 

[USCGDAV]

Based on what I have seen so far, “near perfect scores” are not very likely.

 

[TomT]

With 310 miles over 4 days I am currently at 100/100. Setting following distance to 5, which is what I normally use, helps...

 

[dckiwi]

You only need to get your overall average at or above 99.5

Is there a way to see the non-rounded score? I just see 99%

 

[jebinc]

Is there a way to see the non-rounded score? I just see 99%

No, only by doing the math.