mikes_fsd
Banned
FYI, this is nothing different from Google Maps, Open Street Maps or Apple Maps.
I guess we just live in an HD Map world.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD rewrite will go out on Oct 20 to limited beta
- Thread starter diplomat33
- Start date
FYI, this is nothing different from Google Maps, Open Street Maps or Apple Maps.
I guess we just live in an HD Map world.
no.
you can have non-hd vector maps of course. what makes maps HD is amount of detail recorded there in addition to road directions themselves. What you call "high precision info". so if you list traffic control devices, potholes and other such important road features with high location precision - this is what makes the maps high definition. Hey, creator of the framework that Tesla uses for their maps calls them HD, but Tesla does not
So of course Tesla does not use HD maps by their definition, we already established this.
Now we are trying to establish a common definition for what HD maps really are. So I am just going along with you until you have something somewhat precisely defined but you keep diverting to side topics along the way.
mikes_fsd
Banned
How about, if your solution to autonomy "requires" mapping (HD/precision/etc at this point we're basically talking about Google Maps++) as the foundation for the solution to work (not that it "can do it but we chose not to do it" BS that we've heard all throughout here).
Tesla's "Vision First" solution does not require mapping, mapping is used as an input to the NN and helps with the predictions.
Waymo requires mapping as a foundation - "Before our cars drive in any location, our team builds our own detailed three-dimensional maps that highlight information such as road profiles, curbs and sidewalks, lane markers, crosswalks, traffic lights, stop signs, and other road features."
Each is designed this way from the ground up.
We absolutely can agree that Tesla and Waymo approaches are vastly different. They differ in like every aspect you can think of
Tesla autopilot works without any maps - we know this, I guess you can call this "vision first" even though other sensors (namely radar) also play a role and even override vision at times (see moving iron gate depicted as a car on IC viz as one example) so "sensor suite first" is a less catchy but somewhat more accurate name. It does require nav maps if you want navigation, which is a nobrainer. Performance is improved the more details are recorded on those maps which was already demonstrated.
Now about the Waymo thing - we do not know TOO much. Yes, they have those lidar scans (3d maps as they misleadingly call them), we also know they havily rely on vision and don't blindly trust their maps as was in particular outlined in their relatively recent AMA on reddit: We’re engineering leads at Waymo and we're here to answer your questions on hardware and software development for self-driving technology - Ask Us Anything! : IAmA
This does not really answer the question of what would happen if you drop a (dev) waymo car in some random unmapped area and tell it to go somewhere . I don't think I've seen a straight answer to this from Waymo. It's totally possible the car will declare this to be a "dangerous unsupported operation mode" of course and that's their decision to make, I just don't think this was a clearly defined policy, was it?Waymo AMA said:
Also this:
Waymo AMA said:
So besides initial scan there's continuous "fleet updating" going on (the "expensive to maintain" part that some quote may be not all that expensive)
Lastly I want to add that a lot of the differences probably don't matter wrt the end goal. It's highly unlikely robotaxis will be operating in totally unmapped areas anyway and as long as they have capability to automatically update past the initial scan - the end result (assuming successful completion of the goal at hand) is virtually the same - a robotaxi.
mikes_fsd
Banned
This is the BS every Waymo pusher has been hiding behind and what I referred to "we can do it but we chose not to do it"
Their site and their official twitter account state that maps are the foundation of their system before any car can drive.
A foundation is required, or you can't build stuff on it.
So, by Waymo's words it is a required piece of their stack.
mikes_fsd
Banned
That is not the point of this exercise, does that definition of HD maps satisfy you or not?
(there is a whole thread dedicated to the Waymo bull$#!t -- you can discuss the merits of the cost of their system there)
I agree with it in that context & with Musk that more people would have died & been permanently disabled if not for beta autopilot
I want to see a 10mph hairpin bend, on an unsealed road, with a car traversing the bend in the opposite direction at the exact same time
willow_hiller
Well-Known Member
What a bizarre article. It's almost taking a "man on the street" approach to interviewing:
"Dan Strayer, a tech help desk worker from Waterford, Michigan, says he is not pleased..."
Who is this man? What credentials does he hold? Does he own a Tesla? Why did NBC pick him?
they themselves never claimed this directly in the way you did this "quote".
I am sorry but the message I was replying to did not have any definition, here's what it said:
Is what you are trying to say: "hd maps are a kind of maps that the system cannot work without"? This is a bit murky because:
1. a system that can work without maps cannot have HD maps by this definition
2. This is a bit of a rehash of that other exchange where it was pointed that you still must have nav maps if you actually want to some predefined point. (also by this definition Tesla current maps are HD maps wrt Tesla NoA because NoA does not work unless you have those maps and have an older version that worked fine for navigation, but not NoA)
willow_hiller
Well-Known Member
This raises a bit of a funny conundrum, actually. I agree a L5 vehicle needs to be able to handle a brand new road, but in practice, I don't think it ever will.
Just by virtue of the fact that the L5 vehicle is navigating from one location to another using a street map, it will never choose a brand new unmapped road to travel down. But for that matter, neither would a human driver, unless they were bored, just wandering, or lost. You'd first need to learn about the new road, look it up on a map, and learn where it goes before you'd ever incorporate it into your route. So maybe we can say humans can't drive without an HD map either
We absolutely can agree that Tesla and Waymo approaches are vastly different. They differ in like every aspect you can think of
Tesla autopilot works without any maps - we know this, I guess you can call this "vision first" even though other sensors (namely radar) also play a role and even override vision at times (see moving iron gate depicted as a car on IC viz as one example) so "sensor suite first" is a less catchy but somewhat more accurate name. It does require nav maps if you want navigation, which is a nobrainer. Performance is improved the more details are recorded on those maps which was already demonstrated.
Now about the Waymo thing - we do not know TOO much. Yes, they have those lidar scans (3d maps as they misleadingly call them), we also know they havily rely on vision and don't blindly trust their maps as was in particular outlined in their relatively recent AMA on reddit: We’re engineering leads at Waymo and we're here to answer your questions on hardware and software development for self-driving technology - Ask Us Anything! : IAmA
This does not really answer the question of what would happen if you drop a (dev) waymo car in some random unmapped area and tell it to go somewhere . I don't think I've seen a straight answer to this from Waymo. It's totally possible the car will declare this to be a "dangerous unsupported operation mode" of course and that's their decision to make, I just don't think this was a clearly defined policy, was it?
Also this:
So besides initial scan there's continuous "fleet updating" going on (the "expensive to maintain" part that some quote may be not all that expensive)
Lastly I want to add that a lot of the differences probably don't matter wrt the end goal. It's highly unlikely robotaxis will be operating in totally unmapped areas anyway and as long as they have capability to automatically update past the initial scan - the end result (assuming successful completion of the goal at hand) is virtually the same - a robotaxi.
Its also possible that this is being used as a stepping stone to hasten training? Its possible their short and long term plans are not the same.
their short term plans are absolutely not the same as we can witness for quite some time (Waymo skips the whole L2/L3 steps going straight for L4, Tesla makes own cars, embraces L2). I am not so sure about the long term, but like I said it appears that they are different in anything you can think of (e.g. Waymo does not even promise to license their tech into individual use, though I did not closely watched their Volvo relations development and that might have changed a bit?)
Similar threads
- Question
