Tesla Autopilot maps

[robertvg]

Well you can be tethering over wifi while driving.

I thought that was disabled some updates ago. Works for me only when I'm in park.

 

[DKPowers]

I thought that was disabled some updates ago. Works for me only when I'm in park.

When you switch to drive, yes, it default to 3G/LTE but you can select the wifi again.

 

[DKPowers]

There are reports that auto lane change is now enabled on other roads than highways also, here in Finland. Must mean that the maps are updated.

 

[robertvg]

There are reports that auto lane change is now enabled on other roads than highways also, here in Finland. Must mean that the maps are updated.

Exactly the same here in The Netherlands, at least 3 owners have reported this change since yesterday.
I check one of the tiles in the live branch of a road where auto lane change now seems to work and the file is identical to an older version.
Either some other update or they now use a new url for the tiles ?

 

[StefanSarzio]

I don't think the update is in the tiles. AP1 can do that a long time. Plus it would mean they have to have two sets of tiles. I guess they have some way of sending certain configuration parameters to selected cars, which makes them allow lane-changes on more roads.

 

[Joerg]

I don't think the update is in the tiles. AP1 can do that a long time. Plus it would mean they have to have two sets of tiles. I guess they have some way of sending certain configuration parameters to selected cars, which makes them allow lane-changes on more roads.

Exactly

 

[daktari]

Do we know for certain that both ap1 and AP2-cars use the tiles?
Btw my car in Norway can now also auto lane change where it did not before.
Video in link below.

 

[Pale_Rider]

Do we know for certain that both ap1 and AP2-cars use the tiles?
Btw my car in Norway can now also auto lane change where it did not before.
Video in link below.

Doesn’t look like the video linked properly.

 

[daktari]

Doesn’t look like the video linked properly.

Sorry, meant video link in my signature. But here it is anyway.

 

[AllSeeingEye]

I was informed here that the old approach to the tiled datas was replaced with a new one, so all that you see in the old link is just an old snapshot.
New versions access the data in a different way, yet it is not entirely clear what.

But my friends got out of the car and gave me this file with the data from Norway as an example:

 

[chillaban]

I was informed here that the old approach to the tiled datas was replaced with a new one, so all that you see in the old link is just an old snapshot.
New versions access the data in a different way, yet it is not entirely clear what.

But my friends got out of the car and gave me this file with the data from Norway as an example:

Fascinating....
eu20180518: 2018-05-18 09:31


They're definitely newer. Paging our rooted friends... @BigD0g , @dennis_d, any hints about how to get to these newer tiles?

 

[scottf200]

I was informed here that the old approach to the tiled datas was replaced with a new one, so all that you see in the old link is just an old snapshot.
New versions access the data in a different way, yet it is not entirely clear what.

But my friends got out of the car and gave me this file with the data from Norway as an example:

Seems like 'tile' format at a highlevel is the same as all same options worked.

 

[KyleDay]

Looks like we'll need a new method to determine the new tile URLs

 

[bebbiXpress]

Just took a quick look at that new tile and compared it to the one from the known URL provided by verygreen. The most obvious change is, that the "confidence" value is at 100% for every spline I checked. Even on a very short one crossing another road the confidence rose from 3% to 100% while at the same time F18 changed to true and F27 has the value 3. There are a lot of changes to lane widths and the values F18 to F27 too.

 

[robertvg]

Great news. Date stamp or version information 20180518 seems to explain recent reports of users that auto lane change has been enabled on secondary roads. And the number of splines in this new version is significant higher than in the old one (3346 vs 2279), though still many streets not mapped in this file. Many splines are split into multiple shorter splines.
@DamianXVI the boundaries in the file seems to have changed, can you have a look ?
All we need now is the new url

 

[DamianXVI]

My local map tiles went from 500KB each to around 1000KB each, in the "stage" subdirectory.

Maybe Damian has an idea about the increase in file size.

There are two main reasons why newer tiles (that I checked) are bigger. 1. More splines per tile (more roads mapped). 2. Splines are more detailed (more node points per spline / per mile).

The process of localizing the car position on the spline involves finding the distances between the node points, so maybe the "localizer_optimal" phrase have something to do with the density of spline points.

Seems like 'tile' format at a highlevel is the same as all same options worked.

I looked inside and there are no obvious changes in the file format (only changes in the data that can be seen in the viewer). In case the file structure is updated (more fields are added, without breaking the compatibility) the viewer should show a warning about that after the file is loaded.

@DamianXVI the boundaries in the file seems to have changed, can you have a look ?

Ah, you are talking about the "Bounds" line in the tile information, right? Well this line is only showing a geohash boundaries calculated from the file name. So if you change the tile file name before loading it into the viewer this info will be broken.

 

[strangecosmos]

Where is the source of ground truth redundancy in HD maps if they are compiled using the same neural network that does perception in real time?

Presumably it would be from statistics: if the last 1,000 times cars have passed through an intersection they have detected a stop sign, and now a car doesn’t detect a stop sign there, the car knows the HD map disagrees with its real time perception.

But what about when the environment changes? Some people say HD maps need to be updated daily because the built environment changes so frequently. When the HD map and real time perception NN disagree, how does the car know whether it’s a perception error or a change to the environment?

On the other hand, if the ground truth redundancy comes from human annotation, you’re going to need to collect raw sensor data on every mile of road, and update your HD maps more like once a year than once daily.

If the purpose of HD maps is something other than (or in addition to) sensing redundancy, what is it? Allowing the car to anticipate what’s up ahead on its route? But isn’t an HD map overkill for that?

I’m asking these questions 100% sincerely, not rhetorically — in case that isn’t clear.

 

[lunitiks]

The camera sensors / NNs could be completely or partially blind for a road sign (or something else stationary), e.g. due to fog or rain or light conditions.

When that’s the case, an HD map is obviously good redundancy.

So I anticipate your next question: How would the cameras (NNs) know that they’re being blurred or blinded? Well, I bet Tesla has NNs for that too! Just like they have for raindrop detection on the fisheye NN, they might have «blurry lens» detection as well. @verygreen ?

 

[scottf200]

The camera sensors / NNs could be completely or partially blind for a road sign (or something else stationary), e.g. due to fog or rain or light conditions.

I see stop signs blocked (partially or full) at different times of the year depending on leaves and foliage. Also occasionly due to snow or sun glare.

 

[J1mbo]

I imagine there must be a process to confirm differences. Long term, it might be a UI feature ("report issue") when the car has auto-stopped when it shouldn't, or where the driver has had to take over without being prompted.