Tesla Autopilot maps

[mrkisskiss]

pedal misapplication mitigation...

I remember seeing this mentioned on Chris Lattner’s resume - generally rolled out as part of the update that bumped auto steer up to 90mph.

Presumably, it does exactly what it says in the tin - though I haven’t the cojones to try it... I’d often wondered, if the system sees a car in front of you, and is 100% sure or near enough that it’s a car in front of you (say you’re in traffic)... presumably you can’t just mash the accelerator pedal and slam into the back of the car in front?

 

[daud]

I was downloading and compiling tiles with data from my area and then this [screen shot] happened. I hope Tesla didn't.. @DamianXVI @verygreen any idea?

This specific .tile I did download before, 'new tiles' also return this error page.

 

[BigD0g]

I was downloading and compiling tiles with data from my area and then this [screen shot] happened. I hope Tesla didn't.. @DamianXVI @verygreen any idea?

This specific .tile I did download before, 'new tiles' also return this error page.

I have no problem downloading tiles.

 

[verygreen]

F10&F11 are interesting in this case as well as they show two widths that add up to 4.4m. 4.4m is what is shown as the spline width in the VG snapshot data. Why two width on a single lane off ramp?

If you look further - this ramp splits into two lanes a bit later.

 

[daud]

I have no problem downloading tiles.

False alarm. Big time rookie mistake. Was trying to make myself useful while charging along the highway where I just recorded some AP footage. Just now sitting down at home and saw that I simply forgot to put https in the url.

 

[scottf200]

I just realized this tool I found a while ago is very useful to map 2+ lat/long points from the ADAS Map Viewer tool when looking at particular splines. Especially when you do the beg and end points to see a section in question. This tool does allow you to go down to street level too and it leaves the drop pins on the map when you are there.

Mapping drop pins tool: Map Tool: Map Latitude Longitude Coordinates/Points - Google Maps

Example of how you could just cut-n-paste from the ADAS tool to the mapping drop pin browser page above. Also recall you can right click in the ADAS tool to get the lat/long for two points of a road or 4 points for an area.

a) Current Radar Zone (#1):
Begin: 0 (35.91820786, -84.03115973)
End: 0.232878 (35.91858681, -84.03075976)
Zone Length: 51.40m
P: 97%
Height: 3.250m

b) Spline points (2):
------------------------------------
1: Location: 35.92591364, -84.05748452
Control: -0.00006970, -0.00094815
2: Location: 35.92582014, -84.05843050
Control: -0.00016421, -0.00093197

 

[scottf200]

If you look further - this ramp splits into two lanes a bit later.

Topic 1) I understand. I guess I was trying to figure out why they have it divided into two parts (F10/F11) down where it is one lane. And how did they come up with the split between F10 and F11. I would have guessed on one lane it would be just the road width divided by 2 for both F10/F11. The precision must mean something.

Topic 2) Here is where you were pointing out and the drop pins are the two end points of the spline I'm clicked on. Here is some data from the ADAS Map Viewer. The F10/F11 numbers are even odder here to me (1.74m&5.26m)? In the normal google maps display you can right click and use the measure option from that menu. That is my second picture below. Handy to see comparison values for F10/F11. I did see/realize if you F10+F11 then you get the 7m/22.9ft that the google measuring tool calculates too. I just found it odd that F10 and F11 are so different vs both being around 3.5m.

Topic 3) Also does anyone know/have_guess on what the 'Control' values mean?

Spline points (2): ------------------------------------
1: Location: 35.92591364, -84.05748452 / Control: -0.00006970, -0.00094815
2: Location: 35.92582014, -84.05843050 / Control: -0.00016421, -0.00093197

F10 (Width L?): 1.74 [m] -- 5.7 ft
F11 (Width R?): 5.26 [m] -- 17.2 ft
Sum: 7.0m -- 22.9ft

Right click in google maps on a point to select the measuring tool, then select a second point to get a 'precise' measurement.

 

[daud]

Topic 1) I understand. I guess I was trying to figure out why they have it divided into two parts (F10/F11) down where it is one lane. And how did they come up with the split between F10 and F11. I would have guessed on one lane it would be just the road width divided by 2 for both F10/F11. The precision must mean something.

Topic 2) Here is where you were pointing out and the drop pins are the two end points of the spline I'm clicked on. Here is some data from the ADAS Map Viewer. The F10/F11 numbers are even odder here to me (1.74m&5.26m)? In the normal google maps display you can right click and use the measure option from that menu. That is my second picture below. Handy to see comparison values for F10/F11. I did see/realize if you F10+F11 then you get the 7m/22.9ft that the google measuring tool calculates too. I just found it odd that F10 and F11 are so different vs both being around 3.5m.

Topic 3) Also does anyone know/have_guess on what the 'Control' values mean?

Spline points (2): ------------------------------------
1: Location: 35.92591364, -84.05748452 / Control: -0.00006970, -0.00094815
2: Location: 35.92582014, -84.05843050 / Control: -0.00016421, -0.00093197

F10 (Width L?): 1.74 [m] -- 5.7 ft
F11 (Width R?): 5.26 [m] -- 17.2 ft
Sum: 7.0m -- 22.9ft

Right click in google maps on a point to select the measuring tool, then select a second point to get a 'precise' measurement.

Putting the above coordinates together with coordinates as a result of subtracting control values from the location coordinates looks like this on the map.

Equally the two coordinates linked by their own control values measure both 85,80 meters (281,49 ft) in between.

 

[scottf200]

Putting the above coordinates together with coordinates as a result of subtracting control values from the location coordinates looks like this on the map.
<snip>
Equally the two coordinates linked by their own control values measure both 85,80 meters (281,49 ft) in between.
<snip>

Re: scottf200: "Topic 3) Also does anyone know/have_guess on what the 'Control' values mean?"

In case, it is not very clear all mapping data and concepts are new to me so I really was looking for the layman's explanation of what the 'control' information was. Based on what you said and examples, I scratched my head and did some more google searches. I found this GIS site with some *related* info I think. I could be wrong. I'm guessing now that 'control' is some relative/relation to very accurate known locations so given a second set of coordinates (spline) that you have a offsets to the known gps coordinates which are your control points? (a.k.a. ground control points?)

Some other definitions or ideas that I was guessing were *generally* related to the 'control' values and ideas in these tiles and spline locations.
control point | Definition - Esri Support GIS Dictionary

control point
1) [surveying] An accurately surveyed coordinate location for a physical feature that can be identified on the ground. Control points are used in least-squares adjustments as the basis for improving the spatial accuracy of all other points to which they are connected.
2) [coordinate systems] One of various locations on a paper or digital map that has known coordinates and is used to transform another datasetspatially coincident but in a different coordinate systeminto the coordinate system of the control point. Control points are used in digitizing data from paper maps, in georeferencing both raster and vector data, and in performing spatial adjustment operations such as rubber sheeting.

Another site with some concepts:

Measure Your GCP Center with High Precision GPS
The GPS on your phone, tablet, or drone won't cut it!

You need an extremely precise GPS measurement to create a quality ground control point. To do this you will need a Real Time Kinematic (RTK) or Post Processing Kinematic (PPK) GPS receiver. Traditional examples include the Trimble R family of products (including the Trimble R8 pictured on the right) but recently low-cost alternatives like the Emlid Reach have entered the market.
Link: GCPs for Mapping Contour Lines - Ground Control Points

 

[DamianXVI]

Topic 3) Also does anyone know/have_guess on what the 'Control' values mean?

Putting the above coordinates together with coordinates as a result of subtracting control values from the location coordinates looks like this on the map.

In case, it is not very clear all mapping data and concepts are new to me so I really was looking for the layman's explanation of what the 'control' information was. Based on what you said and examples, I scratched my head and did some more google searches. (...)

You are digging into this too much. All names, terms and concepts in the viewer are put there by me, and I'm not a map/GPS expert. This is the first time I'm working on such topic. To this point there are no actual names defined by Tesla engineers in the viewer, because tile files does not contain ones, only raw values.

"Control" is a term I took from the Bezier splines control points, that can be found for example in Inscape, because the splines in my viewer are drawn using Bezier curves.

Coordinates described as "Control" in Info tab form a relative vector for a node, pointing towards the road direction. You can see those vectors in the viewer when enabling View->Show Control Vectors (they are displayed much shorter to not create too big mess on the map).

Those vectors are used in the viewer to draw road curvature by performing spline smoothing. Here is an example what is happening when those vectors are not used (and a simple line is drawn between nodes), and when those coordinates are used to define Bezier control points:

 

[scottf200]

You are digging into this too much. All names, terms and concepts in the viewer are put there by me, and I'm not a map/GPS expert. This is the first time I'm working on such topic. To this point there are no actual names defined by Tesla engineers in the viewer, because tile files does not contain ones, only raw values.

"Control" is a term I took from the Bezier splines control points, that can be found for example in Inscape, because the splines in my viewer are drawn using Bezier curves.

Coordinates described as "Control" in Info tab form a relative vector for a node, pointing towards the road direction. You can see those vectors in the viewer when enabling View->Show Control Vectors (they are displayed much shorter to not create too big mess on the map).

Those vectors are used in the viewer to draw road curvature by performing spline smoothing. Here is an example what is happening when those vectors are not used (and a simple line is drawn between nodes), and when those coordinates are used to define Bezier control points:

Thank you for the explanation. Your efforts on this tool are so impressive. The use of the bezier control points is fantastic and make perfect sense now. I'm digging because I'm trying to learn new things and I'm assuming you guys know more than I do <grin>. I completely understand you are looking at raw data in bits and likely in hexadecimal displays. Very impressed you could decipher as much as you could. I'm guessing some datatypes like floating point "look" or have a distinct pattern and lengths so you can use them to help define beginning and end points. Anyway I'm off in the weeds.

 

[scottf200]

Ah, now I understand. Yes, it is exactly as you said - there is a byte in the spline structure, that is equal to 1 when the road does NOT have a central divider. At first this line of text was just "F22: Yes/No", later I changed "F22" to "Not Divided" and this is the result. Maybe I should have used the terms "True/False" to better describe what it means. I'll change it when I have time to work on it more.

This confusion may be caused by the language differences. I think it's quite clear by now, that my English skills are not too good. Sorry about that.

Your English is really very good. BTW, I was suggesting to change the label from 'Not Divided' to 'Divided'. My pseudo code was meant to show that the value of 'yes' or 'no' would then be the opposite of the raw data. However, I think from a user interface understanding people have an easier time following positive terms instead of negative terms (i.e. 'Not') because then they do not have to mentally swap/change the meaning in their head. Leave it as you wish as it was just a UI suggestion and I didn't explain it well the first time.

Like you already realized, this list is rather not related to separate measurements, or separate cars. I'm quite sure that this database doesn't contain any individual readouts, only aggregated data. I didn't explained it very well previously, so maybe I'll try to reiterate. I think that this list is showing how the speed is changing along the spline. So when there are three entries in the list, than the first entry shows the observed speed at the beginning of the spline, second in the middle of the spline, and third at the end of the spline. I came to this conclusion after seeing its visualization on the map. It just makes sense, when showing the acceleration and slowdown on highway ramps, slowing down in tight corners, or slowing down when driving thru an intersection: ....

The gradient coloring of the lines for visualization was brilliant and really adds to understanding the data quicker. I see your point on how to interpret the different entries.

 

[Hans Allis]

Wow, so many shiny toys in this thread. I downloaded tile u15zf but am unable to open it using Damian's viewer. It just gives an error saying structure does not contain field #3. Tile u178h works fine, u178n does not. Interesting nonetheless.

 

[AnxietyRanger]

Excellent stuff! I can not begin to thank you guys.

Please keep it up!

 

[NerdUno]

As has been mentioned but is worth repeating, this thread certainly explains why the California drivers appear to be having a better experience with AP2 than some of us further away from the Tesla flagpole.

 

[AnxietyRanger]

As has been mentioned but is worth repeating, this thread certainly explains why the California drivers appear to be having a better experience with AP2 than some of us further away from the Tesla flagpole.

Agreed.

Did I hear someone mention California car?

 

[scottf200]

As has been mentioned but is worth repeating, this thread certainly explains why the California drivers appear to be having a better experience with AP2 than some of us further away from the Tesla flagpole.

To anyone ... Is this just a function of how many Tesla vehicles are sending data back to the Tesla mothership which has input to building the tile data? Assuming a lot of it is automated. Or is there are lot of manual work and decisions on what and how tile data gets update, monitored, and validated. Clearly there must be some serious safety concerns if this data impacts what the car does (i.e. braking on the highway for an overhead sign could be bad)

 

[chillaban]

Agreed.

Did I hear someone mention California car?

It’s worth mentioning that certain other markets like Norway and wherever Buttershrimp is have also been observing superb Autopilot behavior too. I bet they do have the markets they focus first on, but I don’t think it’s just California either.

 

[chillaban]

I’m gonna post this here for lack of a better place:

Just within the last month I’m seeing a dramatic difference in AP2 performance, and I’ve been running either 2017.42 or 2017.44. It’s picking correct lanes after lane breaks, picking correct lanes as new ones form or turning lanes appear, and all “exit diving” on mountainous freeways are gone.

I have to suspect mapping is a huge part of this. I was able to drive 20-30 miles on CA-267 with zero interventions and that is a two lane undivided mountain pass filled with difficult curves and turnouts. Never was able to do that before, not AP1 or AP2, and I don’t think .42 and .44 are different enough to explain this.


The last major problem I’m seeing is that on sharper curves the car drifts towards one side of the lane lines and knows it is doing so. But that’s the only reason I’ve had to intervene and that seems like it’s strictly a control algorithm issue.

 

[AnxietyRanger]

As I said, I reported superior behavior already on .36 on a road populated by a Supercharger vs. another motorway stretch without.

There may be a correlation or heck of a lot of placebo.