FSD V9 First Impressions - General Public Access Seems Way Off

[loquitur]

Some people seem to conflate "autosteer on city streets" with what the current FSD beta allows.
When that feature is released it does not have to include making turns at intersections at all!

What the current FSD testers seem to be doing is really "NOA on city streets" where a non-highway
destination is set. That is where the 90-degree turns happen, and tests much more advanced
features than what autosteer on city streets may provide in a wider release.

With a public release, Tesla can take baby steps. Autosteer on streets may just allow attempts
to drive without lane markings for low-radius curves, allow non-confirmation at green lights,
do nudging around cars with open doors, making an attempt at driving around double-parked cars or trucks, etc.

I do use (public) autosteer on the streets in San Francisco, but only on boulevards with clearly
marked lanes. One becomes quickly aware that it just doesn't work for intersections at crested hills,
streets with hairpin turns even when marked with yellow lines, and comes to a screeching halt in places
like Golden Gate park with many pedestrians and bicycles. It will yield a red hands/takeover now for
someone with an open car door, and would never even think of going around around other stopped cars
in a lane like human drivers would. It won't go around tight curves with parked cars

I think Tesla will only release features gradually, and tempered by ample safety warnings. That
will be fine by me.

 

[rxlawdude]

Nope. That is what Elon said:

"Wide beta maybe with FSD rev 10, definitely with rev 11"

He said wide beta would be either V10 or V11.

He did NOT say in months or maybe longer. Those are your words. Be careful when you allegedly quote to fit your agenda.

With the subscription model, I expect to see much quicker turnarounds, since Tesla would like to maximize that revenue stream.

 

[rxlawdude]

I do use (public) autosteer on the streets in San Francisco, but only on boulevards with clearly
marked lanes. One becomes quickly aware that it just doesn't work for intersections at crested hills,
streets with hairpin turns even when marked with yellow lines, and comes to a screeching halt in places
like Golden Gate park with many pedestrians and bicycles. It will yield a red hands/takeover now for
someone with an open car door, and would never even think of going around around other stopped cars
in a lane like human drivers would. It won't go around tight curves with parked cars

You didn't see the V9 FSD video of Lombard Street?

 

[diplomat33]

He did NOT say in months.

I never said that he did. I said Elon said wide release would be V10 or V11 which is true.

Here is what I said:

It is currently in beta testing (FSD Beta V9) but Elon today said that public release would only come in V10 or V11 so it is several months away at best, maybe longer.

The word "so" indicates that "so it is several months away at best" is my conclusion, not a quote from Elon.

That's your opinion and not sustained by any objective fact. (Of course, past performance is no indication of future results.)

With the subscription model, I expect to see much quicker turnarounds, since Tesla would like to maximize that revenue stream.

Yes, it is my opinion that it will take several months. I stand by my opinion. I seriously doubt that V10 will only be roughly a month after V9 as Elon claims. Yes, past performance can be a good indicator of future results. It took more than a month to do the vision-only rewrite. Linking highway driving, city driving and parking lot driving together under the same software is likely to take more than a month.

The sub model has nothing to do with how quickly V10 will come after V9. Versions will come when the software is ready.

 

[daktari]

I never said that he did. I said Elon said wide release would be V10 or V11 which is true.

Here is what I said:



The word "so" indicates that "so it is several months away at best" is my conclusion, not a quote from Elon.



Yes, it is my opinion that it will take several months. I stand by my opinion. I seriously doubt that V10 will only be roughly a month after V9 as Elon claims.

The sub model has nothing to do with how quickly V10 will come after V9. Versions will come when the software is ready.

v10 in roughly 1 month probably = 3
months probably
v11 in roughly another month = 3 months (+3 months) = 6 months definetly!

 

[linux-works]

as with lots of very complex software projects, the final 2% will take the longest. and that's the safety factor; it already does all the basics but having it do that with enough nines and without damage, that's going to take a long time.

the fact that they are sensor starved and no redundancy where it counts - that's going to hold them back even more.

no manager could ask, sincerely, how much longer its going to take. "we'll know when we get there" is the best anyone can say.

I dont measure it in months, btw.

 

[Twiglett]

I'm one of those idiots/fanboys who upgraded to FSD from EAP because it seemed that the cross country FSD drive was really close, so we would all get it real soon now. That was several years ago.
After seeing the scale of the issues with beta 9 I can't believe people are still buying FSD
Beta 9 FSD might be smoother than previous versions but I can't see you you can be more relaxed knowing it will try to kill you in an instant. You can't take your hand off the wheel for a second in case it tries to swerve into a parked car.
EAP inspires confidence, while current FSD looks improved but scary as hell and very long way from being able to risk with the regular owner population.

 

[Mardak]

Here's some 1-week reviews from those who have used FSD Beta 9:

Dave Lee on Investing

"I was super optimistic with wide release before having the beta. Now I've kind of reversed a bit; I don't know if we're gonna get it by the end of the year… I previously thought that full self driving was at that point where it was a relaxing experience, but after trying it out and then personally talking with a bunch of beta testers, I realized that it's not relaxing. Barely anyone has that overall relaxing; everyone is hyper-vigilant right now because you have to. In order for FSD to get to this place where it's going to be relaxing overall, it's going to take at least a year to a year-and-a-half from now, so maybe end of 2022 or so."

HyperChange

"I though FSD version 9 was going to be a little bit better than what it was mostly because Elon said it was going to be a huge improvement. It wasn't a huge improvement, but it was an improvement. That's very important because on the back-end, the technology moved to version 1 of Tesla Vision, so when you take into account that we dropped radar, moved to this new vision system, pushed out this update, and it performs even better, it's absolutely a huge win for Tesla. This is the progress that we want to see in FSD. This is them moving in the right direction. This is them showing they're leading the charge to commercialize autonomous vehicle to unlock the trillion dollar robo-taxi thesis."

Roger K.

"The car is doing maneuvers much more natural, much more human in most scenarios. The biggest issue for me is this uncertainty when it comes to cross traffic especially cross traffic with higher speed. So for that, I see 2 things: 1) the software needs to be able to detect this better or react better to these 2) probably more important, looking into the future is this communication with the driver to improve that car's mind view to communicate to the driver what the car is going to do next or not going to do."

 

[rxlawdude]

I never said that he did. I said Elon said wide release would be V10 or V11 which is true.

Here is what I said:



The word "so" indicates that "so it is several months away at best" is my conclusion, not a quote from Elon.



Yes, it is my opinion that it will take several months. I stand by my opinion. I seriously doubt that V10 will only be roughly a month after V9 as Elon claims. Yes, past performance can be a good indicator of future results. It took more than a month to do the vision-only rewrite. Linking highway driving, city driving and parking lot driving together under the same software is likely to take more than a month.

The sub model has nothing to do with how quickly V10 will come after V9. Versions will come when the software is ready.

"Elon today said that public release would only come in V10 or V11 so it is several months away at best, maybe longer."

No, you implied an entire quote. Had you said "Elon today said that public release would only come in V10 or V11 so IN MY OPINION it is several months away at best, maybe longer," you would have a point. But you didn't.

Keep it 100.

 

[mark95476]

Tesla is already offering functions and features that no other OEM has or can. They're not stopping and I'm sure they have a lot of additional features in the pipeline. These are going to cost money so everyone should expect fluctuations in the $200/100 monthly subscription.

I think vision-only unblocked a lot of things in their development pipeline so quicker turnarounds are expected.

Tesla removed radar on their high-volume 3/Y line-up and proved their vision-only design and stack was good. It only took about a month to get the IIHS safety features/recommendations on their vision-only (no radar) cars after release. FSD beta v9 is also vision-only and is looking good.

He did NOT say in months or maybe longer. Those are your words. Be careful when you allegedly quote to fit your agenda.

With the subscription model, I expect to see much quicker turnarounds, since Tesla would like to maximize that revenue stream.

Yes, that was impressive.

You didn't see the V9 FSD video of Lombard Street?

 

[stopcrazypp]

Roger K.

"The car is doing maneuvers much more natural, much more human in most scenarios. The biggest issue for me is this uncertainty when it comes to cross traffic especially cross traffic with higher speed. So for that, I see 2 things: 1) the software needs to be able to detect this better or react better to these 2) probably more important, looking into the future is this communication with the driver to improve that car's mind view to communicate to the driver what the car is going to do next or not going to do."

That's a problem with Tesla's approach. Tesla is still operating in the L4/L5 mindset where communication with the driver seems to be more of an afterthought, even though for the foreseeable future it'll be a L2 system. Tesla has a lot of ways they can improve it (including voice prompts, maybe a text based box that shows what the car is planning to do next). Some people may be annoyed by it, but it can be made as a toggleable setting.

Note however that the latency of the car's reaction is not necessarily the same as the car display. AFAIK, the rendering on the car is handled by the Intel Atom processor for the infotainment system, so it is very likely much slower than the actual data processing that the FSD software is using to make its decisions. Personally, I prefer a more communicative interface than spending the rendering resources on pretty graphics.

 

[rxlawdude]

That's a problem with Tesla's approach. Tesla is still operating in the L4/L5 mindset where communication with the driver seems to be more of an afterthought, even though for the foreseeable future it'll be a L2 system. Tesla has a lot of ways they can improve it (including voice prompts, maybe a text based box that shows what the car is planning to do next). Some people may be annoyed by it, but it can be made as a toggleable setting.

Note however that the latency of the car's reaction is not necessarily the same as the car display. AFAIK, the rendering on the car is handled by the Intel Atom processor for the infotainment system, so it is very likely much slower than the actual data processing that the FSD software is using to make its decisions. Personally, I prefer a more communicative interface than spending the rendering resources on pretty graphics.

I think you're on to something re the user interface and how to make driver interaction more integrated with FSD's capabilities (and where FSD simply cannot handle an edge case).

 

[linux-works]

In order for FSD to get to this place where it's going to be relaxing overall, it's going to take at least a year to a year-and-a-half from now, so maybe end of 2022 or so."

like I said, I measure this final few percent of 'done' in years, not months.

even if the curve slows (as I think it will, eventually), the lawmakers may delay the *legal right* to use this tech for an even longer period of time.

what I wish we could get: a US sponsored, vendor-neutral (all access) test ground where its NOT public road, but its like a hollywood set and you can add randomness and stuff to it, UNDER CONTROL.

not impossible to do. biden would buy into this, I would bet. and it would help everyone. in fact, I'd say results should be shared and made a national asset. that way, we (people) would get the tech worked out faster, safer and for EVERYONE.

(too socialist? lol. knew some of you'd say that. this is why we can't have nice things.)

 

[diplomat33]

what I wish we could get: a US sponsored, vendor-neutral (all access) test ground where its NOT public road, but its like a hollywood set and you can add randomness and stuff to it, UNDER CONTROL.

The US already has several such testing grounds. The NHTSA designed 10 test centers as proving grounds for AVs in 2017 but changed their mind in 2018. Kevin O Vincent in his proposal for a framework for AV standards wants the NHTSA to reverse their decision and reapprove those testing centers for AV testing. He writes this on page 27:

SAE has identified 35 different test centers in the United States that have some capability to test autonomous vehicles. In 2017, NHTSA designated 10 of these test centers as official proving grounds for autonomous vehicle testing. The next year, however, NHTSA rescinded the designation of the 10 proving grounds and declared that DOT was not going to pick winners and losers among the test centers. NHTSA needs to reverse course yet again, however, and redesignate a list of centers approved for the testing of AVs. The list of approved test centers should not be limited to the 10 centers previously approved, but all test centers located in the United States should be invited to apply for designation, and NHTSA should approve any center that establishes it has adequate facilities for the testing of AVs. The centers approved by NHTSA must be available for testing by all manufacturers, and they must open to allow public viewing of AV compliance testing (the facilities can be closed to the public at other times while conducting proprietary testing for customers).

Source: https://2uj256fs8px404p3p2l7nvkd-wp...021/05/Kevin-Vincent-Regulatory-Framework.pdf

 

[linux-works]

but there's no - shall we say 'appreciation' - for a shared knowledge base, in order to get to the destination faster and safer. we seem to want to have everyone figure it out, with so much duplicated, wasted and misplaced effort, all on their own.

we could attack this problem in a true distributed form ('fast path'); or just hope some winner strikes it lucky ('slow path').

we do seem doomed to the slow route. pity.

maybe other countries (on their own, or in groups) would make this happen. I just dont see any type of shared cooperation in the US kind of capitalism view. and that's a bug, not a feature, in this case.

 

[Dan D.]

but there's no - shall we say 'appreciation' - for a shared knowledge base, in order to get to the destination faster and safer. we seem to want to have everyone figure it out, with so much duplicated, wasted and misplaced effort, all on their own.

we could attack this problem in a true distributed form ('fast path'); or just hope some winner strikes it lucky ('slow path').

we do seem doomed to the slow route. pity.

maybe other countries (on their own, or in groups) would make this happen. I just dont see any type of shared cooperation in the US kind of capitalism view. and that's a bug, not a feature, in this case.

What you're talking about exists, in a way, in the open source Openpilot system.

openpilot - Wikipedia

en.m.wikipedia.org

comma.ai

Make driving chill. comma.ai has 122 repositories available. Follow their code on GitHub.

github.com

Only works with certain cars, and certain operational design domains (ODD); not Tesla without using a "fork". But if you are feeling Socialist, participate, to the benefit of all.

(Benefit not guaranteed)

 

[thefrog1394]

Features you will likely use, but that have relatively little value

• Auto Lane Change: avoid hearing annoying blim-blom sound from autopilot disconnecting when you change the lane
• Navigate on Autopilot: good reminder for you to take a ramp even if you need to take over often
• Traffic Light and Stop Sign Control: extra pair of eyes reminding of that stop sign / red light.

Features you will try once to conclude they do not work

• Auto Park: Too slow to be useful
• Smart Summon: You will never save a second vs walking to the car

What is on the near horizon

• Navigate on city streets: have been on the near horizon for years, so we do not really know when/if they deliver this.

Really wish auto lane change was available as part of standard AP. AP makes me a left lane hog because in a situation where I'd normally move right and then back left, I have to decide whether the disengage, switch lanes, reengage, disengage, switch lanes, reengage move (and all the beeps that come along with that) are really worth it. Now is that feature alone worth paying 16% of my car's value? Probably not

 

[stopcrazypp]

Really wish auto lane change was available as part of standard AP. AP makes me a left lane hog because in a situation where I'd normally move right and then back left, I have to decide whether the disengage, switch lanes, reengage, disengage, switch lanes, reengage move (and all the beeps that come along with that) are really worth it. Now is that feature alone worth paying 16% of my car's value? Probably not

They likely made it not part of AP because it really is a major feature of EAP/FSD that a lot of people want. I think even a "resume" function (which exists in other cars, where turn signal temporarily deactivates AP, then it turns on again after lane change has been completed) would satisfy a lot of people.

 

[K1Woods]

Also note that the youtube folks have a unique and valued commodity that they can grow their channels with. The longer the time before public release the more the gain. All comments should be taken with a grain of salt to weed through implicit bias. I think determinations on when it goes public should be made by Tesla only. Could you imagine if the original AP was rolled out this way? We would still be waiting…as there are always edge cases - we still see them.

 

[mmyers300]

In the actual Full Self Driving Beta release notes it says, “It may do the wrong thing at the worst time.”
Then Elon tweets that “...there will be unknown issues, so please be paranoid!”

I think I compiled a number of great clips (best of the worst) that would make anyone paranoid about FSD in it's current state. City driving is by far the worst, as expected, but a number of these serious issues occur outside of the city also.
Every time I marvel at FSD I am reminded, by these clips, of how far it still has to go to be viable.

I really love using autopilot on the highway and in heavy stop and go traffic, but there is no way I would use this current version. I know, it's beta, but wow some of these are bad.