Talk:Tesla

Is Tesla using OpenStreetMap data?

So far, I have found the following:

• Twitter user @greentheonly has found Tesla is using Valhalla for routing AND Autopilot decisions. See this Twitter thread and others.
• Autopilot will incorrectly change lanes to "follow route", when ever a motorway exit is mapped in OpenStreetMap as being shaped like a "Y". See this Twitter thread.
• The car will route in ways that are not possible according to the Google map shown on the screen. However, this has been found to happen in places where OpenStreetMap had been corrected two years back, suggesting that if it were routing with Valhalla tiles, they were old tiles.
• There have been many successful cases where improving a parking lot on OpenStreetMap eventually improves the cars ability to navigate them. An update needing only a month or so to take affect suggests that Tesla may be using online routing which is being updated more often than the local Valhalla tiles. See this Tesla Motors Club thread.
• A long thread at Tesla Motors Club attempting to determine the source of Tesla's maps used for Autopilot.
• An old Valhalla fork at Git Hub. One of the members of Tesla fork, noblige, was also an official Valhalla committer.
• An attempt to parse the Valhalla tiles found on a Telsa.
• A discussion on the official Tesla Forum.
• A discussion on the OpenStreetMap Forum.
• Driveways that were marked as private caused a Tesla with Full Self-Driving beta to misroute, until the tags were corrected.

--TreeStryder (talk) 03:29, 29 December 2020 (UTC)

Do you expect your observations to have particular consequences?

--Polarbear w (talk) 17:04, 29 December 2020 (UTC)

For context, I have been contributing to OpenStreetMap for 11 years and have owned a Tesla Model 3 for 2 years. It is obvious that most of the quirks exhibited by Tesla vehicles are caused by them believing their map data; wrong/missing speed limits before a sign is seen, creative routing, late/early/unnecessary lane changes, delayed/missed traffic signal/sign discovery, ignored parking isles, "phantom braking" due to inaccurate mapping or no map data to give the car more confidence in a situation, etc. I hear that Tesla is working to make the car more dependent on its own vision, than on its map data, but it will always need a reliable map. Just like a human would, when driving in a new location (having no mental map). All this to say, if Tesla is using OpenStreetMap data, I would focus some of my limited time toward improvements beneficial to Tesla's full self-driving future.

I have looked for Tesla attributing OpenStreetMap and have not found it yet. This could be because they are distanced from the data. I recently found on a job description "Experience with open source mapping tools (e.g. MapBox) preferred". Valhalla is MIT licensed, no attribution needed. And, it is likely they are getting their Valhalla tiles from a 3rd party.

--TreeStryder (talk) 00:41, 31 December 2020 (UTC)

If it ended up that they were not using OSM data (which I agree is unlikely given all the evidence you've identified, plus others), how would you seek to improve it anyway? OSM is community-sourced, other geodatabases may not be. If an error exists in OSM but not elsewhere, it's easy to fix; if an error exists elsewhere but not in OSM, that's a sign Tesla (and others) should be using OSM. Arlo James Barnes (talk) 08:40, 18 December 2021 (UTC)

I have been meaning to update this conversation for some months. Though @greentheonly on X has confirmed Tesla still maintains Valhalla tiles in their cars, I no longer see evidence that this data is being used. At least, outside of parking lots. It is as if they have changed to a different routing data source. Navigation and planning problems once corrected by editing OpenStreetMap have returned, along with new problems unexplainable by OpenStreetMap history. --TreeStryder (talk) 15:22, 23 June 2024 (UTC)