What will happen within the next 6 1/2 weeks?

[dckiwi]

With approximately six weeks left in 2019, Tesla is running out of time to live up to a specific commitment that has been on the order page since Autonomy Investor Day:

One outcome that seems impossible is that all FSD owners will be enjoying traffic light recognition and 'automatic driving on city streets' by Christmas. For one thing, the HW 2.x -> 3.0 upgrades have not begun in earnest. And presumably HW 3.0 will be required for at least some elements of FSD feature complete.

But rolling out features to HW 3.0 owners before completing most of the retrofits seems like a slap in the face to early FSD buyers (although such a move wouldn't necessarily surprise me).

Just wondering what everyone else thinks is possible/likely.

 

[diplomat33]

With approximately six weeks left in 2019, Tesla is running out of time to live up to a specific commitment that has been on the order page since Autonomy Investor Day:

View attachment 477798

One outcome that seems impossible is that all FSD owners will be enjoying traffic light recognition and 'automatic driving on city streets' by Christmas. For one thing, the HW 2.x -> 3.0 upgrades have not begun in earnest. And presumably HW 3.0 will be required for at least some elements of FSD feature complete.

But rolling out features to HW 3.0 owners before completing most of the retrofits seems like a slap in the face to early FSD buyers (although such a move wouldn't necessarily surprise me).

Just wondering what everyone else thinks is possible/likely.

I agree that at this point it seems impossible that all FSD owners will get both "traffic light recognition" and "automatic city driving" by the end of this year.

I think it was at the last earnings call, that Elon mentioned that Early Access would get these FSD features by the end of this year. So I think Tesla has fallen behind their promised deadline and is now aiming at getting it to EA by end of this year, not to the general public by end of the year. Depending on how it goes with EA, we can probably expect the FSD features to roll out over time in early next year as Tesla also ramps up FSD computer upgrades.

I think we should also expect that FSD owners will not get these features at the same time. Cars that already have HW3 will get them as soon as the OTA update goes out (like they did with traffic cones) and then other FSD owners will get it as they get the FSD upgrade.

 

[DanCar]

With in the next 6 weeks a new early access release will start. Some people will leak:
1. how poor it works
2. others will leak that it is amazing
3. about how regular autopilot has improved

 

[dckiwi]

It seems most people agree that an EA release is coming before end of year with some/all of these features. The big question is whether this release will be limited to AP 3.0. If it is, they better accelerate the retrofit program.

 

[thewishmaster]

I would bet it will be HW3 only since these features are not something that was promised for EAP (enhanced AP) users and thus doesn't "need" to support HW2.5 or below ^

I doubt there will be enough meaningful retrofits by then. On the bright side, by the time we get our retrofits (guessing early next year), these features will be more polished?

 

[EVNow]

In EA Musk said, FC is going well and hopefully they can release it to EAP.

Either Musk has suddenly become conservative or more likely we’ll be lucky to see FC released to EAP this year.

 

[Carnage-]

I think we'll probably see something rolled out to EA users by the end of the year, I'm just hoping I'll get my FSD computer upgrade by then

 

[EVNow]

I think we'll probably see something rolled out to EA users by the end of the year, I'm just hoping I'll get my FSD computer upgrade by then

They seem to be going by VIN (i.e. earlier VINs get updated first). Then there is the geographical progression from west coast of US to east and then to Europe & Asia.

Even with a fairly early VIN (May '18) and on the west coast of US, I don't anticipate upgrade this year.

 

[eli_]

I can't imagine anything getting done between Thanksgiving and the waves of vacationing engineers around December holidays, at least if it's anything like my company, the year is finished by end of this week.

 

[nepenthe]

I can't imagine anything getting done between Thanksgiving and the waves of vacationing engineers around December holidays, at least if it's anything like my company, the year is finished by end of this week.

Presumably, much development effort has been placed on this task (or those DNN's) already and contrary to something like smart summon, automatically taking correct action at a stop light or stop sign is not something you can really release in much of a beta form. As what could they say, "Should stop at stop lights, but be ready to stop on your own just in case!" Or something that would often stop at green lights? This would be quite bad from a potentially more dangerous situation and to the detriment in public opinion as it relates to the regulatory road towards FSD approval. One would not want to release such features without a very heavy amount testing and qa processes. Though I could be way off base and the basic groundwork for the feature, is still being built-out in development now.

But overall, I would wager it will be HW3 only and likely not see EAP release before the year is out - as for how close that is, historically someone like Elon would have been talking about it already.

 

[eli_]

Presumably, much development effort has been placed on this task (or those DNN's) already and contrary to something like smart summon, automatically taking correct action at a stop light or stop sign is not something you can really release in much of a beta form. As what could they say, "Should stop at stop lights, but be ready to stop on your own just in case!" Or something that would often stop at green lights? This would be quite bad from a potentially more dangerous situation and to the detriment in public opinion as it relates to the regulatory road towards FSD approval. One would not want to release such features without a very heavy amount testing and qa processes. Though I could be way off base and the basic groundwork for the feature, is still being built-out in development now.

But overall, I would wager it will be HW3 only and likely not see EAP release before the year is out - as for how close that is, historically someone like Elon would have been talking about it already.

In the most recent Lex Fridman interview with Elon he briefly talked about Smart Summon and FSD and seemed to imply that they will be geocoding traffic light and stop signs in the map data and that scares me. It's basically today's highway NoA extended to city intersections with the same dodgy dependence on map data. Like reading the map data to figure out which lanes have turn restrictions, etc.

I try to imagine putting myself in the shoes of someone letting NoA drive around city streets and navigate complex intersections, and I imagine the stress of monitoring such a system. I'm not a trained safety driver, I don't know if I will be able to deal with it. Unless it was 99.9% reliable I don't think I would use it, it would be too stressful.

 

[diplomat33]

In the most recent Lex Fridman interview with Elon he briefly talked about Smart Summon and FSD and seemed to imply that they will be geocoding traffic light and stop signs in the map data and that scares me. It's basically today's highway NoA extended to city intersections with the same dodgy dependence on map data. Like reading the map data to figure out which lanes have turn restrictions, etc.

I try to imagine putting myself in the shoes of someone letting NoA drive around city streets and navigate complex intersections, and I imagine the stress of monitoring such a system. I'm not a trained safety driver, I don't know if I will be able to deal with it. Unless it was 99.9% reliable I don't think I would use it, it would be too stressful.

The map data is only a guide. The car will still use camera vision to read traffic lights, stops signs, lane markings etc... So for example, when the camera thinks it sees a red light, the car checks the map data to see if there should be a traffic light there. If yes, then it know with higher confidence that the red light is real. If the map data says there is no traffic light in that location, then the car will know that the red light is a false positive. But ultimately, Tesla aims to make the camera vision so reliable that the map data will be irrelevant. Ultimately, the car will navigate completely reliably with just camera vision. Once the camera vision is 99.99999% reliable, it won't be a problem.

 

[Rogue one]

In the most recent Lex Fridman interview with Elon he briefly talked about Smart Summon and FSD and seemed to imply that they will be geocoding traffic light and stop signs in the map data and that scares me. It's basically today's highway NoA extended to city intersections with the same dodgy dependence on map data. Like reading the map data to figure out which lanes have turn restrictions, etc.

I try to imagine putting myself in the shoes of someone letting NoA drive around city streets and navigate complex intersections, and I imagine the stress of monitoring such a system. I'm not a trained safety driver, I don't know if I will be able to deal with it. Unless it was 99.9% reliable I don't think I would use it, it would be too stressful.

Yeah I caught that as well, map data at least where I love is ruffled with errors (mainly incorrect speed limits) couldn’t imagine relying on map data for something as important as stop signs/lights... / :

 

[diplomat33]

Yeah I caught that as well, map data at least where I love is ruffled with errors (mainly incorrect speed limits) couldn’t imagine relying on map data for something as important as stop signs/lights... / :

I doubt that Tesla is relying on map data as the primary source. Tesla uses camera vision. The map data is merely a back up check.

 

[S4WRXTTCS]

Historically speaking Tesla tends to release two major updates a year.

Normally sometimes around the spring/summer, and sometime around the fall/winter. So I'm not expecting anything new aside from bug fixes on the features they recently released.

Now I wouldn't be surprised if there was a lot of activity behind the scenes on collecting sign data with HW3 vehicles. Something operating passively.

So Tesla employees have vacation, but the fleet is put to work.

 

[powertoold]

I doubt that Tesla is relying on map data as the primary source. Tesla uses camera vision. The map data is merely a back up check.

Doesn't really matter though. Elon is falling back on his whole LIDAR/HD maps are crutches statement.

 

[eli_]

I doubt that Tesla is relying on map data as the primary source. Tesla uses camera vision. The map data is merely a back up check.

Yeah I know that's where they want to be, Elon said maps are for "tips and tricks" at Autonomy Day. That's how Highway NoA should work and it doesn't right now, even after a year. But I'm talking about what I think is going into the next build, the first public release of city driving on NoA. And I'm thinking about what the minimum viable thing could possibly be. And what I'm expecting is an extension of what NoA already does (i.e. taking highway forks and selecting lanes via map data) and extending it to support simple intersection traversal.

I think we'll see new vision-based features for sure, like 100% they need super accurate stop line detection, you can't code that into maps at all. That HAS to be vision, but majority of it like lane turn restrictions, intersection topology, right of way, I think they'll try to pull that from the maps. And as a result it's going to need to be super slow and cautious approaching intersections to give the driver time to take control. I think they're spending a bunch of time right now trying to polish that and validate it and make it as safe for folks to use as possible, lots of alert message boxes and new graphics. But I just can't see them going from zero to full vision-based lane markings and road signs, the whole 9 yards in the first public build, that is so difficult to do on a global level. And it's just not how they tend to ship stuff, they try to put out small pieces at a time.

 

[boonedocks]

If it is as bad as summon is I'll wait to use any NOA city driving to catch up to that of a 12 year old driver. A party trick in an empty parking lot is fun to watch but a party trick on the open roads is not for me....or any one that happens to be near my car.

 

[ramonneke]

If the map data says there is no traffic light in that location, then the car will know that the red light is a false positive

That is very important. The false positives and negatives are likely to be returned to HQ to train the neural net as learn that it was not a trafficlight.

 

[mswlogo]

Elon backed off and he said maybe feature complete on “FSD” by end of year to Early Access (Employee Testing). My guess is that might happen by end of Q1 and public might see it by end of Q2. Just a wild ass hunch.