FSD Beta United States Expansion Halt Speculation

[Lowchrg]

Hey folks,

The last time FSD Beta rolled out to new users in the United States was 12/30/21, and today being 3/29/22 means we are almost at the 3 month mark since no new additions to the program.

During this time, users have been continuing to participate in Safety Score which has been fixed to not allow you to clear a drive with a scroll wheel reset. This fix single handedly made driving Tesla vehicles an absolute dread and the most stressful driving experience you could imagine. 2-3 weeks of this is terrible, but 2-3 months is enough to drive anyone absolutely mad.

Happy to see our Canadian friends getting some love but we seem to be all out of answers when it comes to US expansion. Elon has dodged this question on Twitter for weeks on end and that’s never a good sign.

So what do you think is going on? Is the NHTSA blocking further expansion in the US? Is there more upgrades needed (e.g. cabin cam retro for legacy)? Is Tesla just slow?

My personal theory is the NHTSA has suspended Tesla from expanding the program further but if this ever came to light the stock price would plummet. So a strategically placed split soon would distract and pump it enough to bare that blow of bad news once released. I hope I’m wrong though.

 

[clydeiii]

Wouldn’t the NHTSA block be public knowledge?

 

[thmaub]

I don't think this is any sort of regulatory scrutiny. I think they simply have enough users enrolled in the US to gather as much data as they can reasonably process. They expanded to Canada because they now need data to handle the intricacies of that country.

Rolling it wider is probably viewed as a liability because if they are already saturated with data, that just gives more chances for cars to crash or do something that will give them negative press with little benefit.

It's definitely a bummer since I had the beta but lost it due to an AP computer swap, but as a software engineer, I understand their reasoning.

 

[CheckChar362]

There is nothing new for non-beta users for a very long time also. I wish they can give non-beta users something, e.g. go through green lights without confirm.

 

[EVRider-FL]

According to TeslaFi (TeslaFi.com Tesla Model S 3 X Y Data Logger), the latest FSD Beta build (2022.4.5.20) has been deployed recently in both Canada and the US.

 

[S4WRXTTCS]

Personally I think its because there really isn't any development reason to have more people within the same region in the beta program. In fact I hope they realized they made a mistake in having so many people that they'd be unable to answer FSD Beta emails even if they wanted to.

The lack of any customer engagement makes the beta not feel like a beta at all, and the early nature of it make its feel like a developer preview. Something where you can give feedback, but its so early that the developers are too busy fixing things they know are broken to even look at your input.

In fact I think FSD Beta isn't a developer thing at all, and is something Elon wanted to do for marketing reasons. It's a way for Tesla to show owners where its at, and to give them a taste of it.

I feel like it was a mistake.

The Safety Score was a complete mess, and is still a mess. Even if it did work well to measure driving safety they should have only enabled it for vehicles granted into the program where it was the last step.

It didn't last very long until the NHTSA put the brakes on the whole rolling stops thing. Tesla didn't even give themselves a fighting chance at arguing for it because they did so very early.

The lack of customer engagement means people in the beta don't feel like its worth bothering to report anything.

The people in the beta are missing out on bug fixes that happen in the mainline. We have to wait till the next iteration of the FSD Beta. I've had weird slowness/latency in the rearview camera while backing up for months now.

The people outside the beta feel like they're missing out, and this can drive on nutty. It doesn't how much someone with it says "it's hard to use, blah blah blah" because of the psychology of missing out. Maybe FSD beta owners should start a emotional support program for non-FSD Beta owners where we have a FSD Beta meetup. That way people that don't have it can go "Oh, god that was awful". The cops might get called because 20 FSD Beta cars trying to handle a roundabout would cause a traffic jam.

My conclusion is a much smaller, and more engaging beta program would have been better. Where the program was so intense that you got kicked out simply for lack of participation. To satisfy non-beta people some aspects of the beta would be released to the mainline like go on green for example. Then they'd add bing at a 4 way stop when it felt like it was your turn to go.

 

[FelixORiley]

I don't think this is any sort of regulatory scrutiny. I think they simply have enough users enrolled in the US to gather as much data as they can reasonably process. They expanded to Canada because they now need data to handle the intricacies of that country.

Rolling it wider is probably viewed as a liability because if they are already saturated with data, that just gives more chances for cars to crash or do something that will give them negative press with little benefit.

It's definitely a bummer since I had the beta but lost it due to an AP computer swap, but as a software engineer, I understand their reasoning.

I 100% buy this explanation. If it’s true, shame on Tesla for not simply telling everyone in the US that expansion is on indefinite hold. I would be fine with that!

I was genuinely excited about my Tesla. Opted in after about a week post delivery. My whole Tesla ownership has been judged by the invisible safety score, and frankly, it’s awful.

 

[FelixORiley]

According to TeslaFi (TeslaFi.com Tesla Model S 3 X Y Data Logger), the latest FSD Beta build (2022.4.5.20) has been deployed recently in both Canada and the US.

Only deployed to US drivers already in the beta (except for 1 outlier in California).

 

[Lowchrg]

Only deployed to US drivers already in the beta (except for 1 outlier in California).

Fwiw it’s been determined this vehicle originated in California but the owner moved to Canada.

 

[BBTX]

The one outlier in California was on .4.5.4 though, which otherwise only went to Canadian cars to give them the beta opt-in button, therefore it's likely some company test car.

I'd be fine (ok, maybe slightly disappointed) with hearing they're done with beta expansion in the US if they'd actually tell us that's the case instead of stringing us along with pointless Safety Score driving for months.

 

[EVNow]

I wrote this in another thread.

They are rolling out differently this year vs last.

They are likely doing some A/B tests. They rolled 10.9 only to 1/3 rd of cars and left 2/3rd in 10.8.1, for eg.

Similarly they might be leaving a lot of cars in 10.10.2 instead of moving everyone to 10.12.

All this of-course does not explain why they stopped rolling out to new testers in US. May be there effortsare concentrated on rolling out to new testers in Canada.

In any case, hopefully they will restart including new testers soon with the next release - 10.12.

 

[TomT]

It's definitely a bummer since I had the beta but lost it due to an AP computer swap, but as a software engineer, I understand their reasoning.

Same here! It's highly annoying!

 

[nvx1977]

Personally I think its because there really isn't any development reason to have more people within the same region in the beta program. In fact I hope they realized they made a mistake in having so many people that they'd be unable to answer FSD Beta emails even if they wanted to.

The lack of any customer engagement makes the beta not feel like a beta at all, and the early nature of it make its feel like a developer preview. Something where you can give feedback, but its so early that the developers are too busy fixing things they know are broken to even look at your input.

Traditional software bugs tend to be 100% fixed once the issue is identified. That's where a ticketing system and full transparency makes sense. But FSD issues are based on probabilities. Take phantom braking. It's not a "fix this one thing and it goes away." It's more like there are 420 different reasons for a false positive, and Tesla was able to reduce the probability for 69 of those reasons this release, but overall, phantom braking remains an issue. Also, with NN training, some situations will be mutually exclusive, so an improvement in one area might lead to worsening in another, until there is more driving data.

It's just fundamentally way different than traditional software QA and beta testing. I think we are falling into a trap of thinking that they should be similar. The beta's primary purpose is to collect more driving data in more situations. Answering fsdbeta emails was never a priority, and aside from us feeling like it's a poor customer experience to get no feedback from our testing, it's not necessary to engage with each of us to improve the system. We might think that, because we each have our pet peeves and our preconceived notions of how those issues get fixed. But Tesla has their own [clearly controversial] methodology, and we're just going to be perpetually butthurt if we are expecting something otherwise.

 

[Yelobird]

Possible alternate Explanation, their just not done with the next update? Other then forum speculation did they say there will always be updates every X weeks?

 

[n1ghtwaves]

I have also found this to be incredibly frustrating.

I have been paying for a useless subscription for three months now that has offered me nothing - the current software stack of autopilot sucks in comparison to FSD. Summon is incredibly unreliable, auto park is a joke and rarely is ever offered, and lane changes on highway are incredibly slow and indecisive.

It’s starting to reach a point where I want to ask for a refund from Tesla, because at this point I’ve given a $600+ loan to Tesla for the return gift of an IOU.

Oh, did I mention it absolutely sucks to drive my Model S right now because I’m too terrified to ruin my 99 score during rush hour in California?

Who is benefiting right now from even participating? Surely not me - just the tesla FSD twitter and youtube shills.

/rant

 

[Supcom]

I have also found this to be incredibly frustrating.

I have been paying for a useless subscription for three months now that has offered me nothing - the current software stack of autopilot sucks in comparison to FSD. Summon is incredibly unreliable, auto park is a joke and rarely is ever offered, and lane changes on highway are incredibly slow and indecisive.

It’s starting to reach a point where I want to ask for a refund from Tesla, because at this point I’ve given a $600+ loan to Tesla for the return gift of an IOU.

Oh, did I mention it absolutely sucks to drive my Model S right now because I’m too terrified to ruin my 99 score during rush hour in California?

Who is benefiting right now from even participating? Surely not me - just the tesla FSD twitter and youtube shills.

/rant

Obviously, you would be better off cancelling your subscription and waiting for city streets FSD to be publicly released.

 

[S4WRXTTCS]

Traditional software bugs tend to be 100% fixed once the issue is identified. That's where a ticketing system and full transparency makes sense. But FSD issues are based on probabilities. Take phantom braking. It's not a "fix this one thing and it goes away." It's more like there are 420 different reasons for a false positive, and Tesla was able to reduce the probability for 69 of those reasons this release, but overall, phantom braking remains an issue. Also, with NN training, some situations will be mutually exclusive, so an improvement in one area might lead to worsening in another, until there is more driving data.

It's just fundamentally way different than traditional software QA and beta testing. I think we are falling into a trap of thinking that they should be similar. The beta's primary purpose is to collect more driving data in more situations. Answering fsdbeta emails was never a priority, and aside from us feeling like it's a poor customer experience to get no feedback from our testing, it's not necessary to engage with each of us to improve the system. We might think that, because we each have our pet peeves and our preconceived notions of how those issues get fixed. But Tesla has their own [clearly controversial] methodology, and we're just going to be perpetually butthurt if we are expecting something otherwise.

Only some of them are based on probabilities. Other ones are either due to a navigation/mapping issue or some limitation within the SW.

In a lot of cases having the user do a write up of what went wrong in a situation is hugely beneficial. Like a few days ago a TMC member posted about a unique case where a median was in the wrong location, and it tricked the car into thinking a lane was okay to go into. Sure they might be able to deduce that from the reported video if they look at it, but the write up makes it a lot easier. If they never respond then users aren't going to spend time doing write ups.

By far the biggest thing that would benefit my own testing is being able to report Navigation, and maps issues. Just getting those fixed would make the FSD Beta easier to use, and more useful.

The lack of reporting bugs/issues is really biting Tesla in the arse because they're not finding out about things until issues are plaguing the entire fleet.

The auto-lights not going on when its raining during the day is a good example of how some change whether intentional or not
The frustration with preconditioning for fast supercharging is another that's driving a lot of us nuts.
The V11 interface is full of bugs

Overall Tesla needs to implement some way of reporting issues in a tracked manner where customers feel like their input is valued, and is being worked on. The beta differs in that a persons involvement is highly dependent on feeling there is purpose in doing that work. So it makes it even more important to report issues.

It's odd that Tesla's biggest benefit is fleet learning, and yet the customer is removed from it.

 

[OncomingStorm]

Only some of them are based on probabilities. Other ones are either due to a navigation/mapping issue or some limitation within the SW.

In a lot of cases having the user do a write up of what went wrong in a situation is hugely beneficial. Like a few days ago a TMC member posted about a unique case where a median was in the wrong location, and it tricked the car into thinking a lane was okay to go into. Sure they might be able to deduce that from the reported video if they look at it, but the write up makes it a lot easier. If they never respond then users aren't going to spend time doing write ups.

By far the biggest thing that would benefit my own testing is being able to report Navigation, and maps issues. Just getting those fixed would make the FSD Beta easier to use, and more useful.

The lack of reporting bugs/issues is really biting Tesla in the arse because they're not finding out about things until issues are plaguing the entire fleet.

The auto-lights not going on when its raining during the day is a good example of how some change whether intentional or not
The frustration with preconditioning for fast supercharging is another that's driving a lot of us nuts.
The V11 interface is full of bugs

Overall Tesla needs to implement some way of reporting issues in a tracked manner where customers feel like their input is valued, and is being worked on. The beta differs in that a persons involvement is highly dependent on feeling there is purpose in doing that work. So it makes it even more important to report issues.

It's odd that Tesla's biggest benefit is fleet learning, and yet the customer is removed from it.

I agree, a tracking site with what they are currently working on just general updates would be ideal and industry norm for beta testing software really. Just to play devils advocate tho, if their intention is to build an AI that can handle any driving conditions without the need for pre-programming than for that purpose user feedback is all but useless and might actually lead to additional frustration/problems.

that said, yes i think they've dropped the ball, there really should be a way to report things like the headlights and such and really track the dev altogether. i'm guessing they dont want to as they feel it will expose them to further issues with regulators or just bad press / headaches or potentially expose trade secretes and they think the frustration caused by not disclosing anything is the lesser of 2 evils, but the frustration still exists