Openstreetmaps and FSD

[nvx1977]

It's been called out a few times that Tesla may be relying on openstreetmaps for some or all of its mapping data, and anecdotally, I've seen people report FSD behavior that seems to be tied to incorrect mapping data. I encountered an intersection where the car tried to go right through it with no attempt to stop at all. It is 100% reproducible. I suspect a mapping issue.

So I pulled up OSM and the intersection in question. Turns out, the street I'm on before I blow through the intersection (Pleasant St in the screenshot) is marked as "Private." I would interpret this to usually mean like someone's driveway. In this case, the road leads into a condo/townhouse community, and there are custom signs posted stating "residents only." So marking the road "Private" makes sense. But I wonder if that is causing FSD to treat the road like a driveway, and maybe that is why it just goes straight across a road without looking.

I've changed motor vehicle access of that road to all access instead of private. I don't know how often Tesla refreshes map data, or if it's pulling it real-time. I'll test again when I can and report if it makes a difference.

Leaving Pleasant street to continue forward, FSD seems to think Mammoth Rd doesn't exist and doesn't stop to check for traffic before crossing it.

 

[TresLA]

Yikes that's scary. Someone more knowledgeable can give a better answer, but if I recall correctly, our cars download new maps data somewhat infrequently. So it's good that you corrected it on OpenStreetMaps, but your car may not change its behavior until the next time maps data is downloaded (maybe weeks/months between each refresh?)

 

[RTPEV]

Is there a STOP sign on Pleasant Street to indicate that a car approaching the intersection should stop?

Granted, I agree that common sense would dictate that Mammoth Rd should have the right away and Pleasant Street should have an implied stop, but without a sign or map data that tags the intersection in such a way, a computer wouldn't know the difference between this intersection and an intersection with a cross street where the cross street has stop signs, but the current street has the right of way.

While I agree that it's likely a missing STOP tag in the map data, I don't think that it's the fact that the street is marked as PRIVATE that's the problem. PRIVATE roads simply mean that they are not maintained by the municipality, but rather the residents of the street.

 

[EVNow]

Leaving Pleasant street to continue forward, FSD seems to think Mammoth Rd doesn't exist and doesn't stop to check for traffic before crossing it.

I also doubt its because Pleasant is marked private. What is Mammoth Rd marked as ? Can you give the link to it in OSM ?

I'd think Pleasant should be marked "Residential" and Mammoth "Tertiary Highway" ? May be someone more versed with OSM can comment ...

 

[nvx1977]

I also doubt its because Pleasant is marked private. What is Mammoth Rd marked as ? Can you give the link to it in OSM ?

I'd think Pleasant should be marked "Residential" and Mammoth "Tertiary Highway" ? May be someone more versed with OSM can comment ...

OpenStreetMap

OpenStreetMap is a map of the world, created by people like you and free to use under an open license.

www.openstreetmap.org

 

[nvx1977]

Is there a STOP sign on Pleasant Street to indicate that a car approaching the intersection should stop?

Granted, I agree that common sense would dictate that Mammoth Rd should have the right away and Pleasant Street should have an implied stop, but without a sign or map data that tags the intersection in such a way, a computer wouldn't know the difference between this intersection and an intersection with a cross street where the cross street has stop signs, but the current street has the right of way.

While I agree that it's likely a missing STOP tag in the map data, I don't think that it's the fact that the street is marked as PRIVATE that's the problem. PRIVATE roads simply mean that they are not maintained by the municipality, but rather the residents of the street.

hmm good point. Just checked Google street view, and there is no stop sign there. So it could think this is the main street with right of way via vision AI.

Google Maps

Find local businesses, view maps and get driving directions in Google Maps.

www.google.com

So maybe this intersection isn't a mapping issue. But def curious about others who suspect mapping errors. Would love to see the OSM link to see if there's a correlation.

 

[EVNow]

OpenStreetMap

OpenStreetMap is a map of the world, created by people like you and free to use under an open license.

www.openstreetmap.org

I wonder if Tesla thinks Pleasant is under / over Mammoth, since there is no clear junction marked in the map. Ofcourse, there is no bridge mentioned either. Definitely a bug Tesla needs to fix.

 

[FalconFour]

Bump to this! This is a really important point. Things that FSD does are closely related to the map data OSM has. There's a lot of missing info in OSM (a lot of missing stop signs, number/description of lanes, and speed limits, for example), that I'm sure Tesla is trying to make FSD fill-in on the fly. It might really help to go in there and learn to edit things!

I've gone through the onboarding tutorial (much like Waze editing) and made a few submissions of my own for improved metadata. You can't really tell how bad some of the data is until you click the segment, or click a node, and find it's missing all the details.

I think we could make a huge impact by going around and editing areas that throw FSD for issues. It's going to take a lot of effort to fix all the world's map data in this thing! Since it's used for the high-level navigation guidance, no doubt this is going to need to be a fair bit better to help FSD get the job done.

A good getting-started guide, here:

LearnOSM

learnosm.org

 

[nvx1977]

yeah I recently saw a segment of I-93 that was still marked as a 2-lane (per side) highway, even though it got expanded to 4 lanes. This was just a short segment, but it coincided with consistent phantom braking. I changed it to 4 lanes so it's consistent. I was hoping to see an incorrect speed limit, but that was not the case. still, it's a correlation with AP/FSD behavior. So I'll keep monitoring. Wish I knew what the lag time was between update and Tesla picking up on the change.

 

[ItsNotAboutTheMoney]

I wonder if Tesla thinks Pleasant is under / over Mammoth, since there is no clear junction marked in the map. Ofcourse, there is no bridge mentioned either. Definitely a bug Tesla needs to fix.

Don't know how it _was_ but currently no. It's defined as a point with 4 lines intersecting.
But what I do notice is that 3 directions from that point are marked as allowing travel, while the private road isn't and is marked u-turn allowed.
i don't think those directional markers should be there as travel is possible in all 4 directions, it's just that the road is private.
Edit: never mind, those _are_ default behaviors, not explicit.

 

[Kipp]

yeah I recently saw a segment of I-93 that was still marked as a 2-lane (per side) highway, even though it got expanded to 4 lanes. This was just a short segment, but it coincided with consistent phantom braking. I changed it to 4 lanes so it's consistent. I was hoping to see an incorrect speed limit, but that was not the case. still, it's a correlation with AP/FSD behavior. So I'll keep monitoring. Wish I knew what the lag time was between update and Tesla picking up on the change.

It has been almost 2 months since your edit. Have you noticed a difference yet? I'm very curious about this.

 

[nvx1977]

It has been almost 2 months since your edit. Have you noticed a difference yet? I'm very curious about this.

It has not made a difference so far. Confirmed yesterday. The issue is that the car thinks the speed limit changes from 65 to 50mph. I think the speed limit data doesn't come from Open Street Maps. I also checked Tom Tom map data, and the speed limit is consistently defined as 65mph where the car messes up.

 

[JulienW]

It has been almost 2 months since your edit. Have you noticed a difference yet? I'm very curious about this.

No chance of ANY changes until we get a new Navigation Data database download. This happens about 2x a year.

 

[EVNow]

No chance of ANY changes until we get a new Navigation Data database download. This happens about 2x a year.

No - they send small map changes with every release. The question is how do we get their attention and make them send the map changes we need.

They do send a large map release once in a while which should have all the changes ofcourse.

They have to make this proces better - given how much FSD depends on map being correct.

 

[Mardak]

There seems to be map/navigation changes even without a firmware update at least with FSD Beta. I've noticed on 10.5, there's competing navigation planners that might depend on the internet connection.

For a destination with a parking lot, the initial route correctly goes through parking aisles, but after a few seconds, the route changes dramatically ending in the middle of a street with no entrances to the parking lot (but I assume it gets preferred because it ends up closer to the pin minimizing driving through "slow streets").

Even for our corner-lot home, during 10.5, it has switched between wanting to end on the street with our driveway (the street in the address) and ending on the perpendicular street where the pin is just slightly closer to the line.

The map "patch" updates without a full navigational map update could include both road connectivity data that seems to be based on TomTom and other map attributes seemingly from OpenStreetMap, e.g., stop signs, lane counts and turning lanes. And these are separate from the displayed map, which likely updates real-time from Google servers, and these seem to be used for address lookups but not directly for vehicle navigation.

 

[Mr. Michael]

Mapping in my rural area is also very troublesome. Incorrect (always too low) speed limits and blowing through stop signs. I do have the FSD beta (well the whole car is beta really), but while it ”sees” stop signs, it goes right through them unless I stop it. So Tesla Vision isn’t really looking for stop signs, unless maybe cued by the map data. (My map data is from the end of last year.). One very repeatable speed error comes up when it sees a 55 mph sign and figures thats the speed limit, and not 100 yards later, reverts to 35mph, although there are no signs for that - so likely even sign data can be overridden by the (incorrect) map data. Maybe on the plus side, the FSD beta does let you manually set the speed over the limit by more than the 5 mph allowed in autopilot. However FSD is its own can of worms, as it seems to drive on rural highways like its in the city, slowing down for crossroads (when they all have stop signs), and also slowing approaching the crest of hills (no radar to see over them anymore). Would be very nice if we actually had a working process to fix the map errors.

 

[Mardak]

Anybody else have dedicated bus lanes that might be confusing FSD Beta? Looking at the OSM wiki for lanes:bus, I'm pretty sure this is incorrectly tagged:

I believe the lanes should be counting all lanes including turns and bus, so it should be 7 in this case with 4 forward and 3 backward (not 5/2/2). Whereas the current OSM data indicates there's 2 lanes going north and 1 is for bus, and this might be causing FSD Beta to think there's only 1 lane for cars and results in unnecessary lane changes.

Here's a query to find bus lanes: way[highway][~'lanes:bus'~'.']