Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.

Automated driving path(OSM data related)

Sorry about the title, not the best description. Also this is going to talk about Smart Summon a lot but I am not putting it in the Smart Summon master thread because this is about the car figuring out it's driving path in general. Smart Summon just happens to be the primary way to see this in action right now but there are other scenarios where it happens as explained below.

Ok so, I am working on the theory that Tesla is using OSM data as primary data source for driving path in certain instances. The primary ways that Tesla uses this data that I can think of and others have mentioned, for actual driving path, is Smart Summon, and driving on any road where there are no lane lines(at a minimum), or no major lane edge features.

I'll start with the later. So my theory is that if you are in Autosteer and go onto a section of road with no lane lines(required for my theory) and no major lane/road edge markers(possibly applies), that the car uses OSM data exclusively to determine path. If the road isn't tagged at something that is inferred to have two lanes, then the car is going to follow the path of the line as marked in OSM. For example, if the line in OSM is in the center, that is where the car is going to go. If the line gets moved, it will go there.

Smart Summon, same basic theory as above. With people reporting both that the car stays in the center and others saying the car stayed to the right, or left, that seems to be an indication that it is following the line as marked in OSM data. The Parking Aisle tag might not get inferred by Tesla as being two lanes of traffic. This is good when there isn't, but doesn't work as well when there is a wide aisle.

Now if this is true, there are two ways to fix it. First, change the OSM data to something that changes the Tesla behavior. This would be a major change for a lot of things in OSM though. Second, have Tesla's NN not use OSM data as "primary" source data. Analyze using the vision system and supplement and/or check against OSM data. Actually it might be better to use OSM data with the mandatory use of the vision system to adapt to the live reality.

All this is theory on my part at this point but I figured I would start the conversation for those that may have more time to actually look at this and see if it may be true. If the car is using OSM data it might be pretty easy to test by changing OSM data to different things and seeing how the car changes it's routing.
I would guess that the car navigates and drives much in the same way we navigate/drive, breaking the problem down into different levels of solving the problem. (It's architected by people who drive cars, so there is a heavy influence in the way we solve the problem)

1) Strategic
2) Tactical
3) Immediate
4) Reactive
5) Control (accelerate/brake/steer)

Each different level will have a different emphasis on the data that's used. It's also probably not all NN driven. i.e. The strategic planning of getting from A-B on a map is pretty well solved in car navigation systems. The Reactive (don't run into anything) is probably highly NN driven, and it's what's been in place for a long time in the safety systems in the car. The control is probably not NN.

I would guess it's the middle layers, where the work for Summon/City Street Driving is. When do I need to Stop, When do I need to go, Where on the surface do I need to place myself. Which also involves identifying what the boundaries of the playing surface is, and which bits are more preferable to be on.

At this level it's probably blending lots of different information, from map data, model of the world and state, Model of self and capabilities/restrictions. I would guess that this working on a 'World/Self Model' rather than working as a vision/sensor level, whereby it has a model of and behavioral rules of what it expects things in the world to behave like.

I don't believe, it's a simple data-driven approach, it's really an optimization of multiple (conflicting) requirements to come up with a valid solution type problem. (When it can't find a valid solution within it's hard rules, it gives up and the rules say 'stop and let the owner come find me').
Where on the surface do I need to place myself. Which also involves identifying what the boundaries of the playing surface is, and which bits are more preferable to be on.

But that's the rub. Some people have seen the car stay on the right side of a parking aisle during smart summon, some the left, some the middle. The question of my post is why. If it is effectively random as the car determines on its own using the NN then fine, NN needs to be adjusted.

I don't believe, it's a simple data-driven approach

I never said any of this is simple, nor to I equate "data-driven" as simple. Everything is data and there is no simple about it...My post theorizes that OSM data itself(or the interpretation thereof), if relied on by the car, might be the issue. This is something that potentially could be tested real world by the normal(or abnormal) enthusiasts.

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.