Tesla Motors and street-mapping services

[Crowded Mind]

Classic Model S owner here.

As Musk explained, every Tesla Model S produced has been recording its GPS position and relaying that data back to Tesla HQ, making a highly-detailed map of every road a Tesla has driven on.

Unlike other GPS systems, this information gives a degree of accuracy which allows Tesla to discern discrete lanes in each road, as well as real-world tracks for things such as intersection turns and crossings. As a slide from Tesla’s presentation shows, Tesla has even collected data from parking lots, showing the individual layout of parking spaces and the routes in between them.

Even when the car is out of signal range of an LTE cell tower, Musk said the cars cache recorded GPS data and upload them when they next communicate with Tesla’s servers.

You still count!

 

[replicant]

I'm curious to know what data exactly is in the database. Google's Map Truth makes the project really daunting.

 

[Cobbler]

TM is now building "extreme high-res" street-maps because (almost) every Tesla on the road is recording the exact road/lanes/signs/.... to a big Tesla-cloud? So the more Tesla's on the road, the better the autopilot function will get.

Could they plan to sell access to this Tesla-cloud to other car-manufacturers so they can use it for their autopilot software?

 

[wdolson]

TM is now building "extreme high-res" street-maps because (almost) every Tesla on the road is recording the exact road/lanes/signs/.... to a big Tesla-cloud? So the more Tesla's on the road, the better the autopilot function will get.

Could they plan to sell access to this Tesla-cloud to other car-manufacturers so they can use it for their autopilot software?

If they are recording this data, they are getting very good maps of a few areas, but leaving out 90% of the streets in the world. There are lots of Teslas in the Bay Area, but very few in Oklahoma. Even in the Bay Area you're only getting coverage on streets Teslas have driven down. That would include most major streets, but very patchy on residential streets.

Google has probably gathered much better data than Tesla could with a shotgun approach like this. Google is deliberately driving down every street in the world recording everything with a car that is specially equipped for mapping including GPS.

 

[Crowded Mind]

TM is now building "extreme high-res" street-maps because (almost) every Tesla on the road is recording the exact road/lanes/signs/.... to a big Tesla-cloud? So the more Tesla's on the road, the better the autopilot function will get.

Could they plan to sell access to this Tesla-cloud to other car-manufacturers so they can use it for their autopilot software?

Musk mentioned in that call that this may be something they would consider in the future.

 

[Drax7]

Mobileye computer vision algorithms and processor are the brains behind the tesla
assisted driving. They are the artificial intelligence behind teslas system.
Most of the rest I read here is incorrect .

 

[CLK350]

If they are recording this data, they are getting very good maps of a few areas, but leaving out 90% of the streets in the world. There are lots of Teslas in the Bay Area, but very few in Oklahoma. Even in the Bay Area you're only getting coverage on streets Teslas have driven down. That would include most major streets, but very patchy on residential streets.

Google has probably gathered much better data than Tesla could with a shotgun approach like this. Google is deliberately driving down every street in the world recording everything with a car that is specially equipped for mapping including GPS.

Update FYI - This is now Sept 2020 and we are at the cusp of having Internet access everywhere on planet earth via SpaceX's Starlink project. So it is now a matter of months - this objection will no longer be valid.

 

[replicant]

Update FYI - This is now Sept 2020 and we are at the cusp of having Internet access everywhere on planet earth via SpaceX's Starlink project. So it is now a matter of months - this objection will no longer be valid.

https://twitter.com/jackiewattles/status/1138574349231886339

 

[CLK350]

https://twitter.com/jackiewattles/status/1138574349231886339

Ah - indeed - it may take a few more months+ - this would happen in two step: first the major towns along the internet un-serviced areas roads will get internet via a dish to Starlink, and *they* will retransmit locally via 900 Mhz locally; then relays will be strewn alongside roads ..

 

[adiggs]

I don't think that's how they work. They both need a) a map (it'd better be able to cache it, for going through tunnels or backwoods Maine), so the car knows what route to take, b) eyes/radar on the immediate surroundings.

You can't do only (b) and expect to navigate anywhere. You can't do only (a) and expect the car to get through traffic, cyclists, pedestrians, obstacles, road works, etc.

If you want an autonomous car, the map is 100% essential. Is Tesla OK with using someone else's? Are there competitive suppliers should the one turn bad?

For such a critical piece, should Tesla in fact be in full control by owning it? (Like BMW/Audi/Daimler are doing - Audi, BMW and Daimler Near Deal to Buy Nokia Mapping Service - WSJ)

I'm not sure that the map is 100% essential for an autonomous car, though that depends on exactly what you want the car to use the map for. If the use is something closer to today's usage for route planning (help me get from here to there), then yeah - a map is essential.

But if the map is being used moment to moment to successfully navigate the streets, then I think it's a lot fuzzier.


Mobileye is doing street mapping, and as I understand it, those detailed maps are available (probably downloaded as tiles) as an assist to the equipment and in-car calculations used to drive the car. So it's not the only input, but it is A input.

As I understand it, Tesla is relying solely on in-car sensors and calculations to do the driving, and of course, Elon has observed that humans drive using only our eyes, so a first principles view of things suggests that cars should be able to do the same. Except of course that most of us can identify corner cases where we don't see how Tesla gets from here to there.

A simple example I've been experiencing recently - I've got a twisty up/down a mountain road I drive reasonably often. I think AP could drive most of it today (though I haven't tried). There are corners that I know from other experiences that it simply can't get around. Think 15 mph hairpin type of corners. I think what's going on there is that the corner is so sharp that the front facing cameras can't see where the car needs to go.

This by the way sounds very similar to the problem the car will need to be able to solve to make right hand turns at intersections, so maybe this is easier than I think. The point though is that detailed street mapping could be helpful in that situation to help the car understand where the road goes, even though the car goes through a short period where the front facing cameras can't see.


I don't see any way that a map centric approach (vs a map-help approach), with onboard sensors to avoid hitting people or cars has a chance of working. But I'm not an expert or somebody that works in the field, so these are my hypotheses.

 

[CLK350]

Another possibility is for Tesla to send Starlink-equipped dish antennas cars (probably Cybertrucks) on these un mapped roads; they would stop every once in a while and transmit their recorded positions and images, so Tesla HQ can construct the missing maps to be used in static mode, aided by vision cams to navigate theses newly charted roads.
Am sure they have it planned one way or another (too late to buy TomTom, but who knows)