Elon: "Feature complete for full self driving this year"

[diplomat33]

Thanks for all that!

So, the car cannot distinguish between a green light and a red light. It just knows that there's a light and it will stop unless/until the driver says go. That's not encouraging.

No. The car can detect green and red lights. It shows them on the driver display correctly. But I guess if it is not 99.999999% reliable yet so Tesla is playing it safe by making it stop at all lights just in case it is wrong while they gather more fleet data. But like I said, Tesla will undoubtedly lift this restriction when they get confirmation from fleet data that the feature is reliable enough.

Level 3: I have to be in the driver's seat and awake but I need not pay attention. The car will alert me with a reasonable amount of time (at least half a minute) when I need to take over.

Yes. That sounds about right.

Level 4: I can take a nap in the back seat. The car will pull over safely and stop in a safe place and wake me up if I need to take over.

Yes but I would add one amendment:

L4 can completely take you to your destination while you sleep. It just depends on whether your destination is still in the car's ODD. So for example, Waymo has L4 that is geofenced. If you are traveling within the geofenced area, the L4 can take you to your destination without needing to wake you up.

Of course, what I'd really like is a car that does what a chauffeur would: Take me where I want to go and wake me when we get there.

Again, that could be L4 or L5 depending on the car's ODD.

The current L2 system has even more immediate needs for the driver to take over and has no camera-based driver monitoring.

Yes and I think Tesla should have a driver facing camera system for L2.

L3 specifically does NOT require the driver to be paying attention to the road at all

He just has to be conscious and able to take over eventually (rather than immediately on an L2 situation where he's technically the one actually driving all the time).

So L3 I can be reading a book while the car drives, and if it beeps at me appropriately I can put down my book and prepare to take over. L3 does not require me to be paying attention, at all, to the driving task- just that I be available to take over eventually.

Yes, I know what L3 is. That is why I wrote this:

L3 means that the driver does not need to pay attention in certain conditions but does need to take over when the car requests it.

The driver does need to be able to take over when the car asks them too so what happens if the driver is not able to take over when asked to? How does the car know if the driver is complying with the take over request? Without a driver facing camera, you are relying on just the torque system to detect if the driver really has taken over when requested or not. You can't have a safe L3 if the car asks the driver to take over but has no way of reliably knowing if the driver has complied or not. And yes, I am aware that a L3 car could be programmed to gradually slow down and stop if it thinks that the driver has not responded to a take over request in the allotted time. But without a driver facing camera, the take over process would not be very reliable IMO.

Basically, the driver facing camera plays a different role in L2 and L3. In L2, it monitors that the driver is paying attention to the road during operation. In L3, it does not care if the driver is paying attention to the road during operation but it does check if the driver responds to a takeover request.

 

[daniel]

Here's what they promised for FSD prior to March 2019...

They made a lot of promises. You quoted one. Elsewhere they promised that the car could be sent off to pick up the kids, be a robotaxi, or whatever. L4 can stop safely and tell you you must take over. This would be unacceptable in a robotaxi or if the car is bringing your kids home from soccer practice.

What we know in retrospect was that at the time of making those promises it was nothing but a pipe dream. They had no idea whatsoever how to accomplish it. They just believed, in principle, that a car could be made to drive itself. They designed it with the hardware (sensors and computer) that they thought should be adequate. And made a promise they really had no idea if they could keep or not. One of their promises was that the car had all the needed hardware, which turned out to be mistaken and they had to retro-fit the computers. I believed then, and still believe now, that my car does not have adequate sensors. And now that they clearly cannot fulfill their promise, they are refusing to admit that they took money under false premises. They owe at least an apology and an offer of a refund with interest to the early buyers of "FSD." They just quietly changed their promise to something they can claim to have fulfilled with a set of Level 2 features that work pretty well or that sort of work when the situation is good.

...ish?

It only works to a distance of about 200 feet. It's a neat party trick. And if I am lucky enough to get a VERY close parking spot at work then I manage to be JUST inside the range to use it and summon it to the covered area in front of the front door so I can get in dry when it's raining- but otherwise it's not super useful.

(same feature is available to folks who got EAP back when that was a thing)

So, in my opinion, it's not justified to say that they have delivered on the promise of smart summon. They said it would do those things. Not that it "would do them some of the time when conditions are just right." Yeah, a lawyer could make a case that if it works half the time in small enough parking lots that "the car does what was promised." But I think that's sleazy.

No. The car can detect green and red lights. It shows them on the driver display correctly. But I guess if it is not 99.999999% reliable yet so Tesla is playing it safe by making it stop at all lights just in case it is wrong while they gather more fleet data. But like I said, Tesla will undoubtedly lift this restriction when they get confirmation from fleet data that the feature is reliable enough.

Thanks for the clarification.

So, let's say you're approaching a green light, and the car starts to slow down, so you tap the accelerator to tell it the light is green, but then the light turns yellow while you're still far enough away to stop safely: Does the car recognize the yellow and initiate a stop again, or at this point do you have to completely take over and stop the car?

As for Tesla lifting the restriction, that's a similar issue to raising the rating on the FSD features from L2 (where it's debatable whether it's justified to even call it FSD since the driver is at all times responsible for the car's actions and must be ready to intervene without notice) to L3 or L4, where you can finally leave the driving to the car. Note that after several years, everything the car does under EAP are still called beta features.

Given what we've seen so far, I don't think we're going to see Level 3 and above on any 2020 and earlier Teslas during the average time that these cars remain in service.

 

[diplomat33]

So, let's say you're approaching a green light, and the car starts to slow down, so you tap the accelerator to tell it the light is green, but then the light turns yellow while you're still far enough away to stop safely: Does the car recognize the yellow and initiate a stop again, or at this point do you have to completely take over and stop the car?

No, I believe it will stop the car. It does recognize yellow lights.

As for Tesla lifting the restriction, that's a similar issue to raising the rating on the FSD features from L2 (where it's debatable whether it's justified to even call it FSD since the driver is at all times responsible for the car's actions and must be ready to intervene without notice) to L3 or L4, where you can finally leave the driving to the car. Note that after several years, everything the car does under EAP are still called beta features.

Given what we've seen so far, I don't think we're going to see Level 3 and above on any 2020 and earlier Teslas during the average time that these cars remain in service.

When I say "lifting the restriction", I simply mean having the car automatically go through green lights without requiring any input from the driver. It would still be L2. I am not talking about upgrading to L3 or above.

It is similar to what Tesla has done with NOA where they required a stalk confirmation for lane changes at first and then later made the stalk confirmation optional but NOA is still L2.

I agree that we will probably not see L3 or above on current Tesla hardware.

 

[ChrML]

I feel Tesla's promise for self-driving is a completely unneccessary company risk. They could've called it Software updates towards self-driving capability.

And simply stated that if you buy that package, your car will get free software updates towards an end goal of full self-driving over the coming years, but end result depends on successful validation and regulatory approval.

They would still have sold just as many cars, and people would be impressed by the updates. What they actually did was setting the bar really high for themselves, hence no update will impress anyone.

 

[Soda Popinski]

I can say from personal experience that the feature appears to be reliable. I've not had any big issues with it on my roads. It has stopped at stop signs every time. It has stopped every time for red lights and goes on green when I tap the stalk. It has correctly stopped when the light is yellow and turns red.

So far for me, it's never missed a stop, but has got 2 false positives. One was a stop sign meant for people driving out of a parking lot, but was angled so it could have been seen from the main road. It's since been taken down, so not an issue anymore. The other was more interesting - there was a big rig truck with 3 brake lights in a horizontal row on top of the tanker being hauled. There were about 3 or 4 cars between me and the truck, and my car started giving me the signal that it's coming up to a traffic control / intersection, displaying a horizontal traffic light where the truck's brake lights were. This was on the freeway at an interchange, in stop and go traffic. Obviously, I never reached the "intersection" since it was also moving forward at the same rate I was.

 

[diplomat33]

So far for me, it's never missed a stop, but has got 2 false positives. One was a stop sign meant for people driving out of a parking lot, but was angled so it could have been seen from the main road. It's since been taken down, so not an issue anymore. The other was more interesting - there was a big rig truck with 3 brake lights in a horizontal row on top of the tanker being hauled. There were about 3 or 4 cars between me and the truck, and my car started giving me the signal that it's coming up to a traffic control / intersection, displaying a horizontal traffic light where the truck's brake lights were. This was on the freeway at an interchange, in stop and go traffic. Obviously, I never reached the "intersection" since it was also moving forward at the same rate I was.

I did get one "double stop sign" on the driver display but it did not affect the stopping feature.

 

[Soda Popinski]

No, I believe it will stop the car. It does recognize yellow lights.

I can confirm this. Often times, I'll be approaching a green, get the warning 600 feet out, tap to confirm, but it turns yellow to red, and the car will stop. Under the exact same circumstances where it remains green, the car will continue through the intersection.

 

[Knightshade]

They made a lot of promises. You quoted one. Elsewhere they promised that the car could be sent off to pick up the kids, be a robotaxi, or whatever.

Where, during the actual purchase process for FSD, did they promise anybody that?

Random crap Elon said on an investor call is not a "promise", legally.

L4 can stop safely and tell you you must take over. This would be unacceptable in a robotaxi

Weird- because that's literally how the robotaxis Waymo is currently running work.

There's a remote human in a call center who can help the car out if it gets into something outside its operational domain.

or if the car is bringing your kids home from soccer practice.

Again- citation required they ever "promised" your car could do that.

One of their promises was that the car had all the needed hardware, which turned out to be mistaken and they had to retro-fit the computers.

Indeed. And have stated multiple times since that any future needed retrofits would also be done for free.

(this first came up when HW2.5 became known, well before HW3 was a thing)

They just quietly changed their promise to something they can claim to have fulfilled with a set of Level 2 features that work pretty well or that sort of work when the situation is good.

Again- the FSD they are selling under the new promises isn't the same product.

So, in my opinion, it's not justified to say that they have delivered on the promise of smart summon. They said it would do those things. Not that it "would do them some of the time when conditions are just right."

Not sure how you get that they didn't deliver on smart summon.

smart summon (for older buyers) is part of EAP, not FSD.

And the only thing promised during purchase was
https://www.teslarati.com/wp-content/uploads/2016/10/Tesla-enhanced-autopilot-upgrade.jpg
"be summoned to and from your garage"


Which arguably BASIC summon could already do.


And even when they described smart summon elsewhere it's described as:

"your car will navigate more complex environments and parking spaces, maneuvering around objects as necessary to come find you in a parking lot"

And in the example picture shown it basically goes like 50 feet around a corner and that's it


Smart summon does all that.

It doesn't say "will navigate ALL levels of complex..." just more than it did before.

 

[daniel]

No, I believe it will stop the car. It does recognize yellow lights.

I can confirm this. Often times, I'll be approaching a green, get the warning 600 feet out, tap to confirm, but it turns yellow to red, and the car will stop. Under the exact same circumstances where it remains green, the car will continue through the intersection.

Thank you both. This is encouraging.

Where, during the actual purchase process for FSD, did they promise anybody that?

Elon's continual talking about "these" cars (the ones being made at the time) being able to do that stuff might not be legally binding, but a promise from the CEO of a company is certainly ethically binding. It was repeated over and over that these cars would be able to drive themselves without a human present. I knew before I placed my order that I didn't want the FSD package because I didn't believe they could deliver. So I have no idea what would have been in the purchase contract.

As for Waymo's call center, I am not taking about a company operating a fleet of robotaxis. I'm talking about Elon's promise that we could operate our cars as robotaxis.

Also, the promise of updated hardware becomes a broken promise as soon as they realize that FSD requires hardware that it's not possible to retrofit into the older cars.

I feel Tesla's promise for self-driving is a completely unneccessary company risk. They could've called it Software updates towards self-driving capability.

And simply stated that if you buy that package, your car will get free software updates towards an end goal of full self-driving over the coming years, but end result depends on successful validation and regulatory approval.

They would still have sold just as many cars, and people would be impressed by the updates. What they actually did was setting the bar really high for themselves, hence no update will impress anyone.

Absolutely! They could have said "We hope to develop FSD some day. Buy this package and you will get software features as they become available." That's a lot like what they're selling now, which is a tacit admission that their earlier promises cannot be fulfilled.

 

[Knightshade]

Elon's continual talking about "these" cars (the ones being made at the time) being able to do that stuff might not be legally binding, but a promise from the CEO of a company is certainly ethically binding.

In the context of forward looking statements during investor meetings- it's literally not

From the Wiki on Forward Looking Statements-

In United States business law, a forward-looking statement or safe harbor statement is a statement that cannot sustain itself as merely a historical fact. A forward-looking statement predicts, projects, or uses future events as expectations or possibilities. These statements can often be misleading, as they can be mistaken for factual statements, while they are actually speculation.


These are statements that are specifically not binding at all. They're simply future possibilties the company hopes will happen- NOT any sort of binding promise of any kind.

As for Waymo's call center, I am not taking about a company operating a fleet of robotaxis.

I mean, you specifically said "This would be unacceptable in a robotaxi"

Apparently now it is acceptable if a "company" is doing it?

Why does it matter if a company has 5 of these running around, or if one rich private person who owns 5 of em does?

How's that relevant to L4 vs L5 being "acceptable"?

I'm talking about Elon's promise that we could operate our cars as robotaxis.

You mean his forward looking statement that's explicitly NOT considered a factual statement?

Also, the promise of updated hardware becomes a broken promise as soon as they realize that FSD requires hardware that it's not possible to retrofit into the older cars.

Since you've got a working crystal ball, can I get the next winning set of powerball #s real quick?

 

[daniel]

Again, I'm not talking about statements made during investor meetings. I'm talking about statements made to the public.

I apologize for being unclear about robotaxis: Elon said that individual, ordinary car owners could use their car as a robotaxi when they did not need to be using it themselves. I misspoke if I gave the impression that I was saying that fleet robotaxis would not be practical at that level of capability. I was referring to Elon's oft-repeated promise that if an ordinary person bought a Tesla for his/her own personal use, he/she could, for example, have the car take her/him to work and then go off and earn money as a robotaxi, and then pick him/her up from work, and then go off and earn money again as a robotaxi until needed again. And of course, take a break from robotaxi to take the kids where they needed to go. This would not be feasible if the car periodically stopped and needed a human to deal with an issue.

Elon clearly over-promised. I don't see how anybody could fail to see this. In the past he's always over-promised on delivery times. With FSD he over-promised on performance. The downgrading of the promise shows this pretty clearly.

And, no, I don't have a crystal ball, but I'll bet dollars to doughnuts that the cars built before Tesla downgraded their definition of "FSD" will never reach unsupervised L4 operation because that would require hardware that for practical reasons cannot be put in them.

 

[Knightshade]

Again, I'm not talking about statements made during investor meetings. I'm talking about statements made to the public.

....like?

"random stuff dude says on podcasts" or "random stuff dude posts to twitter at 3AM" aren't necessarily official company press releases after all.

I apologize for being unclear about robotaxis: Elon said that individual, ordinary car owners could use their car as a robotaxi when they did not need to be using it themselves. I misspoke if I gave the impression that I was saying that fleet robotaxis would not be practical at that level of capability. I was referring to Elon's oft-repeated promise that if an ordinary person bought a Tesla for his/her own personal use, he/she could, for example, have the car take her/him to work and then go off and earn money as a robotaxi, and then pick him/her up from work, and then go off and earn money again as a robotaxi until needed again.

And of course where he announced that was INVESTOR autonomy day, and specifically phrased it as a forward-looking statement, not a promised fact.

And when asked about the idea since then basically gave some version of they still hope to do that sometime in the future so again- forward looking statement, not a specific promise of a deliverable.

Again if you've got some public statement from the company NOT in the context of a forward-looking statement, but an actual legal promise it'll drive your kids to soccer, please provide a source.


LOTS of companies do this sort of thing- they just don't tend to get nearly as much press as Tesla.

And, no, I don't have a crystal ball, but I'll bet dollars to doughnuts that the cars built before Tesla downgraded their definition of "FSD" will never reach unsupervised L4 operation because that would require hardware that for practical reasons cannot be put in them.

What hardware, specifically, and why would it not be practical to install it?

 

[daniel]

P.S. I really hope I'm wrong, and if Tesla sells a car capable of L3 operation or above, on any public road that I would otherwise drive on here on Maui, in any weather that I would normally go out in, and with on average fewer than, say, two disengagements per week of my normal driving (I don't put on a lot of miles here) then I will either pay for the upgrade or (more likely) trade in my car for that one. (More likely the trade-in since that would probably get me the features sooner, and I expect that my car will be ready for trade-in by then anyway. But even if they issue the upgrade today.)

Note: The car need not be capable of driving on the Highway to Hana because I'd never go there because I get carsick. It also need not be capable of driving on any of the unimproved jeep trails because I don't drive on those either.

 

[daniel]

What hardware, specifically, and why would it not be practical to install it?

Sensors. And because of cost. Maybe even a later generation of computer that cannot talk to the rest of the car. But more likely sensors. I don't believe my car has adequate sensors for reliable unsupervised L4 operation. And I expect that sensors would be more difficult to retrofit than a new computer, as long as the computer is compatible.

Note that Tesla expected the HW 2 computer to be adequate and because of the promise (which you seem to think was not a promise) they've had to kludge in new computers. It will be much harder if they have to kludge in lidar and more radar and more cameras.

 

[Knightshade]

Sensors. And because of cost.

What sensors, and how much do they cost?

Because again it seems like you are just guessing.

I don't disagree it's POSSIBLE L4 or L5 will turn out to require $10,000 more sensors. But it's also POSSIBLE it'll end up costing $1000 in sensors and it'll be a lot cheaper to do that then send everyone refunds.

Maybe even a later generation of computer that cannot talk to the rest of the car.

HW3 was explicitly designed to be direct swap into the car in place of HW2. There's no reason to suspect that wouldn't be true of HW4 either (which we know has already been under development for over a year- and from a MFG perspective would make SENSE that you can easily swap one in for the other).

Note that Tesla expected the HW 2 computer to be adequate and because of the promise (which you seem to think was not a promise) they've had to kludge in new computers. It will be much harder if they have to kludge in lidar and more radar and more cameras.

IIRC Green noted HW3 goes from a 100mb ethernet port to 1GB... just spitballing but if they DID find they needed more sensors there's currently no inputs for on the computer they could just have an aux computer that gathers that data and sends it to main computer via gig ethernet, which I'd expect to be plenty fast enough to handle that.

 

[P3dStealth]

I just purchased FSD last night to test for 48hrs.

Honestly I am quite impressed. I drove to work today 65 miles from Suffolk county long island to Manhattan. It works really well. It detects all the traffic lights even when I have 10 visible in a row I couldn't get it to make a mistake. If you get a green to yellow it starts stopping even after you hit the stalk or gas. You can't make it run a red or yellow. The lane changes were good. It's way better than everyone made it seem.

It does have a serious problem with lane changes when people aren't letting you in. In NY you have to pretty much just force your way in and this car will almost stop on the highway if it cant get to an exit.

I also had one phantom brake so that was weird.

I haven't tried summon or autopark yet.

Seems like the car nags more.

Also if you are in the right lane it will let everyone on the ramp go before you and almost stop lol.

I do think this could be mostly hands off of they get turns on city streets. I could have been mostly hands off today except for that.

 

[powertoold]

Tesla would never release a level 3 feature. It's either level 5 or nothing, based on their approach. If it's good enough for level 3, then it's good enough for level 5. Based on my understanding.

 

[diplomat33]

If it's good enough for level 3, then it's good enough for level 5. Based on my understanding.

I am not sure that is correct. L3 means that the car can only drive autonomously within a very specific ODD and also cannot always handle its own fall-back and therefore may need to request that the driver take over. L5 means that that car can handle the fall-back itself for a full, unrestricted ODD. So, L3 will be more limited than L5. So L3 does not mean that you are good enough for L5.

 

[willow_hiller]

Tesla would never release a level 3 feature. It's either level 5 or nothing, based on their approach. If it's good enough for level 3, then it's good enough for level 5. Based on my understanding.

Actually that begs an interesting question. We keep talking about the race to L5, but can you call a vehicle L5 until it can drive on any public road? Including roads outside of the US?

Several companies are testing autonomous vehicles in single cities right now, and they might eventually run in several states, but who is going to be the first to achieve international L5? As far as I know, Tesla is the only company operating vehicles with autonomous features across multiple countries simultaneously.

 

[P3dStealth]

I will be driving home soon. I will let you know how it goes later tonight. I consider Manhattan to eastern long island about as hard as it will get.

Its not that far fetched to think Tesla could eliminate the times I took control on the way in. I pretty much took control to force a merge and turn on NYC streets.

I don't think the taxi thing is going to happen soon but they do say you have to use their app and assuming some profit goes to Tesla what if they do what waymo does and they have a monitor watching remotely?