here's an interesting question that occurred to my do you think we will, in the general release see the " active dot" representation which by its very nature will be very jumpy and noise filled or maybe an intermediate representation like what the traffic lines, turn markers , cars, cones and trashcans are rendered? it is a lot of processing just getting the dots and their color registrations let alone making that into a more uniform view , but I could see that ( the white line/. grey object) being the future renderings versus such an active display for the beta testers. unfortunately it is a lot more code going from raw result to the cleaner smoother view. a couple of revs back if you remember the "dancing car shapes" before they were able to edge register front and back . once we all feel it is working and reliably seeing intersections and activity ( now it just suppresses that at a certain distance out, like when cars turn right or left ) , I could see a smoother view as smooth as the way it shows the traffic lights , speed limit signs and trash cans moving by. seeing that as the representation of an intersection would be slick! ( one can dream....)
-
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
Cheburashka
Active Member
powertoold
Active Member
I have a suspicion that the 2019 demo used HD maps, considering the stability and detail of the islands and intersections.
EVNow
Well-Known Member
If I exclude all posts with "HD Map" in this thread, it would be only 1 page long, instead of 109.
UI is improving rather quickly. They can probably change the threshholds for things to visualize, teach the neural network to be less jumpy, zoom a bit more cleverly, make the boxes into 3D mesh objects, lower the opacity etc and make it look pretty clean. I think it’s a winner, it will amaze people and help winning the trust that the car is really freaking good at seeing things.
Daniel in SD
(supervised)
Better yet, they could get rid of the visualizations and people could keep their eyes on the road. Trust in a system that they explicitly tell you not to trust seems like a bad thing! Seeing things is not the most difficult part of FSD, understanding things is.
Agreed, but at least its stopped the endless posts about when Tesla will fit LIDAR to their cars
.
>> Agreed, but at least its stopped the endless posts about when Tesla will fit LIDAR to their cars
They just morphed into how seeing is not really a problem, understanding is ...
The universe did not and will not run out of arm-chair critics.
They just morphed into how seeing is not really a problem, understanding is ...
The universe did not and will not run out of arm-chair critics.
Daniel in SD
(supervised)
No one is going to watch the visualization for the many thousands of hours necessary to determine the accuracy of the perception. I think a lot of people underestimate the reliability required.Agreed, but at least its stopped the endless posts about when Tesla will fit LIDAR to their cars
How would we determine if Tesla needs to add lidar to their cars?
Thousands of hours? You can do this in minutes with the current wide release. From the random scale (i.e. the "banana" effect to object position when you pass them and detection switches from front to side cams), to the ~3 second lag... but IMO this is not down to the lack of LIDAR: it is down to the UI being a low-priority bolt-on.
I don't personally have an issue with the current beta UI. It does the job. I'm sure they will polish it up over at some point but I don't think that's necessary for a wider release. The UI is a supplement to my own eyes in order to ensure that the car sees what I see. I don't care so much how it displays what it sees as long as I can decipher what it sees.
EVNow
Well-Known Member
Well, you may not care - but Elon does. So, we'll get a much polished UI - whenever we get the FDS (beta).
Cheburashka
Active Member
Yes, it’s the latest FSD beta, but there’s some debate on whether it’s the one Elon has referred to with various functionality changes.
Kim had a great vid on her Twitter of the new update handling a roundabout like a champ. Eager to see Brandon test the new software on his weird ass roundabouts in Sac.
Last edited:
DDotJ
Member
Looks like @greentheonly found his way into the AP/FSD Dev UI and settings.
Dev UI has a toggle for an FSD Default preset and a Production Default preset. I assume the FSD Defaults is what FSD Beta is running.
There's a lot of settings, found a reference to "Selfie Driver Monitor", but it doesn't seem to be enabled by default yet for the FSD or Production presets.
and another toggle enabled for FSD called "California Stop" which I assume is the slow rolling past of stop signs seen in FSD Beta videos
And not to add kindle to the bonfire of debate about HD maps again, but take this as you will. Looks like Tesla is using mapped intersections, stop signs, traffic lights, lanes, stoplines, and direction of traffic to aid FSD.
Dev UI has a toggle for an FSD Default preset and a Production Default preset. I assume the FSD Defaults is what FSD Beta is running.
There's a lot of settings, found a reference to "Selfie Driver Monitor", but it doesn't seem to be enabled by default yet for the FSD or Production presets.
and another toggle enabled for FSD called "California Stop" which I assume is the slow rolling past of stop signs seen in FSD Beta videos
And not to add kindle to the bonfire of debate about HD maps again, but take this as you will. Looks like Tesla is using mapped intersections, stop signs, traffic lights, lanes, stoplines, and direction of traffic to aid FSD.
Last edited:
Similar threads
- Question
