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

FSD rewrite will go out on Oct 20 to limited beta

This site may earn commission on affiliate links.
I could understand if someone thinks FSD is SD map and vision fusion. We know Tesla uses maps for stop sign anticipation (at least for now). I could imagine that the FSD neural net uses what it knows about intersections from an SD map to "dream" about future intersections, like how Google's DeepDream can start with a photograph and imagine patterns out of it. But again that's only used for anticipating things yet unseen by the vision system. If there's anything incorrect in the output's assumptions about the future intersection, it would be corrected by the time the car can see it.

But in all likelihood, the majority of these "I can't see the intersection up ahead, how is FSD predicting it?? They must use HD maps!" phenomenon can probably be explained by the wide angle forward facing cameras being situated higher than any photographs taken inside the car.
 
I could understand if someone thinks FSD is SD map and vision fusion. We know Tesla uses maps for stop sign anticipation (at least for now).
Oh, Tesla uses maps ...
In the same video that Karpathy (February presentation) bashes HD maps he says, "we do use maps, they are not HD maps".

But note, we are not talking about geo-coded locations of stop signs and traffic controls, we are talking about "centimeter level accuracy"
You can argue this is not "centimeter-level" HD maps all you want, but entire road markings are premapped in this example.
 
  • Like
Reactions: WattPwr and hh83917
Waymo / Google defined HD maps as centimeter accurate 3D maps. Tesla uses maps, and someone could call them HD maps, but they don't meet Google's definition.
You are going to sidetrack this into some bs technicallity.
The argument green made on twitter is pretty clear.
He is saying that Tesla lied and is using HD maps with lane markings.

I am willing to bet $100 that Tesla is not using HD maps or even maps with "lane markings" in any end-user owned Tesla.
 
No.
You are able to recognize situations you are not able to handle and either stop and/or turn back and/or take a detour. L5 does the same.
Yet this is means it is not capable of driving on ALL roads. There are roads it can't drive on.
And it is just fine.

Yes, I mention in my reply that L5 has to be able to pull over, stop or take a detour when a road is not driveable.

L5 has to be able to do all roads as long as they are driveable. In other words, L5 cannot say "it's driveable but I don't do that type of road". For example, L5 cannot be geofenced or limited to just highways.
 
Yeah, I am disappointed. FSD Beta is not available in Indiana where I live. :(

So where do I go to sign up for this FSD Beta? :p
Muahahahaha.... NC got it...
but not me yet. :(:(

upload_2020-10-25_9-48-16.png
 
Last edited:
Maps are a good thing. Humans use them. When we drive in an area we are unfamiliar with, we tend to drive slower.
yes...
Oh, Tesla uses maps ...
In the same video that Karpathy (February presentation) bashes HD maps he says, "we do use maps, they are not HD maps".

But note, we are not talking about geo-coded locations of stop signs and traffic controls, we are talking about "centimeter level accuracy"
 
As you have already discovered, there is at least one tester in NC. Here is a video from there.
I love the part "I don't have anything in the Nav, it's turning right!" and the car turns right...

I don't know where you're going, where are you going car!??

Code:
https://youtu.be/gi-K0i465xM

This is my new FAVORITE video of FSD.
Nothing in the navigation, the car is literally deciding where to take him. The owner seems just as amused as I am!
 
I think Tesla are using some form of basic maps mainly for navigation but also with some minor additions such as speed limit, traffic lights, stop signs and maybe even “false radar detections”. A map then basically is a long list of road lane objects where every object has a list of which lane they connect to(both forward and sides), to help the car decide when to change lane. Also with some extra attribute such as a few GPS cooridinates along the way for visualization in the display and as a heuristic for the search algorithms etc.

A HD map includes much higher level of detail, here the driveable area needs to be calculatable with some precision.

Then there are a lot of different layers that can be added. Waymo uses a layer for localization which is a 2D bitmap of Lidar reflectivity for a fast and efficient localization algorithm. Using that they can say where the car is lon/lat/rotation. Knowing the positiong they can figure out the driveable area in the road layout map layer using the same coordinate system.

Imo Tesla are not trying to localize themselves in a map with centimeter level precision. But using their neural network they figure out how many lanes they have to choose from and which lane they are in and then using the map they decide which lane they should transition to.

Imo what Green is seeing is not the map, it’s most likely the neural network’s output of how it thinks the world looks like. It just happens to be so good that it looks like a map even when there are obstructions for a hand held camera in the backseat.
 
That map shows only the handful of states where FSD Beta has been located because users who live there are posting videos.

It is not all states where FSD Beta is available. Only a subset of FSD Beta users (we have no idea what percentage AFAIK) have permission to post video -- the rest are under full NDAs.

So the map shows the minimum number of states with FSD Beta -- it could be all 50 states plus DC so far as we know.
 
Maybe it is a mismatch on what we define as FSD?

I don’t really see that L4 will be coming from FSD in the next year or five.

I think it’s entirely possible it could be pretty good and go a couple or maybe even 10 miles without interventions in complicated situations relatively soon, and be perhaps 10x or 100x better than it is currently, in the next year, though!

I just don’t see how that is all that useful (for me). I mean, maybe it is useful for some people but I am not sure how safe it might be.

To be clear, this is fine as far as I am concerned. I just wanted better freeway driving and the better active safety features, which will likely be possible with HW3 at some point. I don’t ever expect my car to drive itself anywhere point to point while I read a book or play with my phone (even if that is legal somehow), and that is fine!

I think there’s a better than 1 in 4 chance of the HW3 computer saving me from doing 12k in damage to property, or bodily injury, in the next 5 years. (Assuming I am not using city FSD of course.) I guess that doesn’t say much about my driving? Probably the chances are lower than that. But I think that with reasonable further progress, the tradeoff here might be somewhat rational. It sucks to be in an accident. Especially an at fault one.

I couldn't agree more with this. I know I've said this a bunch of times different places on this forum, but it's no less of an amazing accomplishment if FSD is still a Level 2 (or a very limited Level 3) system! Autosteer on city streets and highways with an average of one disengagement or fewer on the average commute would have been unthinkable for $10k 5 years ago, and now it's close to being a reality. With all respect to @diplomat33 , I agree with what you are saying about the relative strengths and weaknesses of Waymo and Cruise versus Tesla, but to me Level 5 isn't the point. Having a solid Level 2 system has enormous value for the average driver both for convenience and safety. I think we shouldn't conflate FSD and Level 5 autonomous driving. So much of the current angst about FSD was just caused by Tesla promising Level 5 back in 2016-2018. Without those crazy optimistic promises, I think most people would be incredibly happy with where we will be soon with a solid Level 2.