Where is the enhanced autopilot?

[Peteski]

Those that paid for FSD got the advanced sensor suite, but it's capabilities have not been utilized with the currently available software.

Are you saying that cars ordered with FSD get a different sensor suite to those without? I thought they all had the same sensors coming off the line and FSD was just a software update? Or maybe a hardware update too if required at the time.

 

[Lipsyte]

Are you saying that cars ordered with FSD get a different sensor suite to those without?

They all have the same hardwares

 

[_jal_]

Oddly enough, a couple of months ago the software knew to decelerate before a sharp turn on my drive home. Now it doesn't, and I have to take control before that corner unless I'm following a car. Otherwise it wouldn't be pretty.

The problem isn't just the regressions. It's the fact that the behavior changes without explanation or warning.

YES! Didn't Tesla have to program the car to slow down into those turns? Wouldn't some unit test throw up an error that the car isn't slowing down on that stretch now? I mean somebody coded "Slow down going into these (whatever "these" are) turns." And now they don't, so shouldn't that be an error during automated testing? Blah

 

[electracity]

This is zero percent the customer's fault and don't try to call him out on it. As a customer, I don't give up my rights of fair representation if I do zero research on a product. It's not on the customer to filter and fact check everything or even anything that a sales consultants says. The onus is 100% on a sales consultant to present factual information especially as they are Tesla employees and they represent the company.

ELON never used the word "PROMISE". Follow the rules, man. No magic word so no complaining. Besides, with fleet learned and narrow AI FSD is just around the corner so the missing features in EAP will work. No worries.

You need to read more posts in the investor forums and get you head straight.

 

[DinoJuice]

YES! Didn't Tesla have to program the car to slow down into those turns? Wouldn't some unit test throw up an error that the car isn't slowing down on that stretch now? I mean somebody coded "Slow down going into these (whatever "these" are) turns." And now they don't, so shouldn't that be an error during automated testing? Blah

I guess it depends on how much they're using AI/NN to control driving. Elon has hinted in the past that as they're finding some "drivers" are better than others. I think this hints at something like a GAN having at least some control over driving inputs (perhaps with more traditional control-logic to handle exceptions or define limits). In such cases, there's no code to unit test and it's actually quite hard to debug what's happening as the NNs make decisions. The focus then turns to tuning the training data. The whole field of AI has many challenges to solve with regards to debugability.

Nick

 

[_jal_]

I guess it depends on how much they're using AI/NN to control driving. Elon has hinted in the past that as they're finding some "drivers" are better than others. I think this hints at something like a GAN having at least some control over driving inputs (perhaps with more traditional control-logic to handle exceptions or define limits). In such cases, there's no code to unit test and it's actually quite hard to debug what's happening as the NNs make decisions. The focus then turns to tuning the training data. The whole field of AI has many challenges to solve with regards to debugability.

Nick

I really don't know enough to assert anything, but I'd think they could simulate driving by feeding data to the system inputs for basically any state. Like in The Matrix. I could see how that could get more complicated if the state was actually based on previous system outputs that could change from model to model. Actually, I think that's what the simulated driving programs (the video game driving program) is supposed to tease out.

Still, they've got these beta testers. Shouldn't they have them test specific paths in the real world? At some point they have to be able to test specific reactions; they can't just turn it loose without running some specific tests.

 

[JohnnyG]

I guess it depends on how much they're using AI/NN to control driving. Elon has hinted in the past that as they're finding some "drivers" are better than others. I think this hints at something like a GAN having at least some control over driving inputs (perhaps with more traditional control-logic to handle exceptions or define limits). In such cases, there's no code to unit test and it's actually quite hard to debug what's happening as the NNs make decisions. The focus then turns to tuning the training data. The whole field of AI has many challenges to solve with regards to debugability.

Nick

I guess my question on this, as an AP1 driver, is how much of this AI/NN is currently or will ever be used on AP1 cars?

The reason I ask that question is; in this exact example you guys are referencing, I experience this constantly with my AP1 car.

There is a sharp corner a mile from my house that I go through at least twice a day. Most times it slows down gradually and makes the turn at a reasonable speed. Other times it doesn't even attempt to slow down at all and barrels through it. Some times it gets the angle right and makes it all the way through the turn. Most times it looses it's mind in the apex of the corner and dives to the inside.

And yes, it's behavior does seem to have changed, depending on firmware version. Note I said changed; not progressively gotten better.

 

[croman]

I'm not sure it's AI/NN dependant. That's hard coded driving algo.

 

[whitex]

If Elon was telling the truth about almost being able to do the FSD coast to coast drive, and I have no reason not to believe him, then that would suggest that Tesla is actually pretty close to having FSD. This would lend support to the notion that Tesla feels that they are close enough to finishing FSD that they might as well just wait a bit longer until FSD is done before releasing the new EAP features.

There are plenty of reasons not to believe Elon, primarily the list of things he promised in the past and never delivered. As for FSD being almost ready, don't you think if it was remotely true, they would just put that software on cars, disable all functionality except for lane-keeping, but at least it would have the safety features of being aware of all cars and people around it (including cross traffic) and for example not be hitting parked cars. Unless of course by "almost finished" FSD you mean it still hits parked cars, still can't see cars around or cross teafic - basically currently EAP with additional features which work even worse than current EAP features.

 

[diplomat33]

As for FSD being almost ready, don't you think if it was remotely true, they would just put that software on cars, disable all functionality except for lane-keeping, but at least it would have the safety features of being aware of all cars and people around it (including cross traffic) and for example not be hitting parked cars.

No. We don't know what the missing piece was that prevented the coast to coast demo. Maybe the missing piece prevented them from releasing the FSD software in limited "autopilot" mode. The software is incredibly complex.

Unless of course by "almost finished" FSD you mean it still hits parked cars, still can't see cars around or cross teafic - basically currently EAP with additional features which work even worse than current EAP features.

Please, don't be sarcastic. If the FSD demo was going to be that bad, Elon would never have said that it was almost finished. Clearly, by Elon's words, the FSD is much better than that. The most reasonable interpretation of Elon's words is that the FSD is good, just not good enough for the public. Tesla wants to wait until FSD is better than "good enough".

 

[MorrisonHiker]

No. We don't know what the missing piece was that prevented the coast to coast demo. Maybe the missing piece prevented them from releasing the FSD software in limited "autopilot" mode. The software is incredibly complex.



Please, don't be sarcastic. If the FSD demo was going to be that bad, Elon would never have said that it was almost finished. Clearly, by Elon's words, the FSD is much better than that. The most reasonable interpretation of Elon's words is that the FSD is good, just not good enough for the public. Tesla wants to wait until FSD is better than "good enough".

I always imagined they were using some Nvidia tech and software combined with the new hardware. When the relationship between Tesla and Nvidia fell apart, they could no longer do that and had to develop FSD themselves from scratch. I don't know if that's the case or not but it makes sense to me.

 

[MarcusMaximus]

There are plenty of reasons not to believe Elon, primarily the list of things he promised in the past and never delivered. As for FSD being almost ready, don't you think if it was remotely true, they would just put that software on cars, disable all functionality except for lane-keeping, but at least it would have the safety features of being aware of all cars and people around it (including cross traffic) and for example not be hitting parked cars. Unless of course by "almost finished" FSD you mean it still hits parked cars, still can't see cars around or cross teafic - basically currently EAP with additional features which work even worse than current EAP features.

I’d note this assumes that FSD can be pared down to just the EAP feature set. If, for example, FSD is being trained as a single large network going from input images to vehicle controls, it may not be possible at all.

 

[croman]

No. We don't know what the missing piece was that prevented the coast to coast demo. Maybe the missing piece prevented them from releasing the FSD software in limited "autopilot" mode. The software is incredibly complex.



Please, don't be sarcastic. If the FSD demo was going to be that bad, Elon would never have said that it was almost finished. Clearly, by Elon's words, the FSD is much better than that. The most reasonable interpretation of Elon's words is that the FSD is good, just not good enough for the public. Tesla wants to wait until FSD is better than "good enough".

This is absolute drivel. You have no idea about the software, whether it even exists, much less about its complexity. You also have less than no knowledge of FSD, Tesla's developmental state, or Elon's state of mind.

I miss the unicorn and @verygreen.

 

[diplomat33]

This is absolute drivel. You have no idea about the software, whether it even exists, much less about its complexity. You also have less than no knowledge of FSD, Tesla's developmental state, or Elon's state of mind.

I'm just expressing my opinion. And we are all in the same boat when it comes to knowing the state of Tesla's FSD. None of us have any direct info on the state of FSD unless we work for the Tesla FSD team, in which case we would probably not be allowed to post here. My opinion is no less valid that someone with no knowledge saying that FSD is a scam and will never happen.

 

[JohnnyG]

Sooo, I'm just gonna throw this out there... (I feel like a mouse that just walked into a room full of pissed-off cats)

I kind of feel like maybe we're all taking out our aggression on each other, when really we've not done anything to each other to cause this. Maybe we're all just a little uptight at the status of EAP/FSD and our varying options wouldn't matter much if the product was already here and functioning. Instead, everyone would be arm-in-arm dancing, drinking, praising the majestic powers of EAP/FSD.

So come on guys... can't we all just get along?

 

[diplomat33]

Sooo, I'm just gonna throw this out there... (I feel like a mouse that just walked into a room full of pissed-off cats)

I kind of feel like maybe we're all taking out our aggression on each other, when really we've not done anything to each other to cause this. Maybe we're all just a little uptight at the status of EAP/FSD and our varying options wouldn't matter much if the product was already here and functioning. Instead, everyone would be arm-in-arm dancing, drinking, praising the majestic powers of EAP/FSD.

So come on guys... can't we all just get along?

I agree.

I do find it interesting how FSD-skeptic this forum seems to be. Every time I try to express an optimistic or positive take on FSD, I get slammed for being ignorant and dumb.

 

[Altes]

With the class action lawsuit, Tesla should be issuing small refunds to EAP owners, due to the delay in providing EAP.

What would have been better - allowing EAP/FSD owners a one-time option to get a full refund for their EAP/FSD purchase if they've changed their minds and have grown tired of waiting for these options to be implemented.

EAP provides value now, even though it doesn't achieve all EAP goals. On a recent road trip, AutoSteer was engaged almost the entire way.

We also believe we'll start seeing some benefit in the next 12 months from the FSD purchase with the AP software using the additional cameras, and AP working better under more conditions (compared to AP1).

While we're disappointed in the longer-than-expected implementation time for EAP/FSD, we included EAP/FSD in the configuration of our new Model X (which we hope to get next month).

WOOHOO CLASS ACTION LAWSUIT WON BY "THE PEOPLE" ELON MUST PAY!!! WAY TO STICK IT TO THE MAN!!! VICTORY FOR THE LITTLE PEOPLE!!

.....oh...wait...$233 for each litigant....nevermind

 

[_jal_]

I agree.

I do find it interesting how FSD-skeptic this forum seems to be.

You have to appreciate that there’re a lot of people who paid very real money for a feature that hasn’t had an iota of proof of development. There really isn’t anything anyone can hold up as proof. And a lot of them 18 months ago had the same confidence that you now have that it’s release was imminent. After a while you get to the point where you feel like the Great Pumpkin is a joke and that you’ve been cheated out of tricks and treats.

 

[The Duke]

You have to appreciate that there’re a lot of people who paid very real money for a feature that hasn’t had an iota of proof of development.

Strange, I drove Nicki onto 680 South at Fremont this morning. I hit that little lever twice and relaxed for 30 miles until the 280 - 880 interchange like I do 2-3 times a week nowadays. She could not do it safely 8 months ago. Construction, people cutting her off and speed changes from 85 mph to stopped were handled just fine today like every day lately. I am not going to tell her she "hasn’t had an iota of proof of development" as it might make her revert back to the darting, weaving and phantom stops of 8 months ago.

 

[diplomat33]

You have to appreciate that there’re a lot of people who paid very real money for a feature that hasn’t had an iota of proof of development. There really isn’t anything anyone can hold up as proof. And a lot of them 18 months ago had the same confidence that you now have that it’s release was imminent. After a while you get to the point where you feel like the Great Pumpkin is a joke and that you’ve been cheated out of tricks and treats.

I do appreciate that people paid a lot of money many months ago for features that are not available yet. But I don't think it is fair to say there is not "an iota of proof of development". We know that Tesla has invested a lot of money in hiring FSD experts and in adding new hardware (AP2) to all new cars coming off the production line. Would they have done that if they were not serious? We've seen AP software updates that have vastly improved AP's auto steer. Karpathy who is chief of AI tweeted that he rewrote the neural net. Tesla asks owners to consent to them collecting videos for the purpose of FSD. We got that leak that I think Teslarati reported a few months ago that shows someone driving a Model S in FSD development mode. We got the report of Tesla upgrading the CPU (AP2.5). I could go on. There is certainly proof that Tesla is working on FSD. I grant you that we have not seen a lot of progress but there is certainly evidence that Tesla is working on it.