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

Seeing the world in autopilot, part deux

This site may earn commission on affiliate links.
It seems to be an ongoing controversy whether without lidar HW2 Teslas are capable of estimating depth accurately. Would you or anyone you know be able to run a proper trial of Autopilot’s depth estimation metadata against some ground truth (like a long measuring tape or a laser distance measurer)? Ideally including as one of the tests from a distance as far away as you can accurately measure with your ground truth instrument.

I think if Autopilot always comes in at within 20 cm or 8 inches of ground truth, that’s hands-down a win for Tesla. A bit worse would probably be fine, but humans only seem to manage about 20 cm of accuracy longitudinally when parking.

Humans do better on lateral accuracy: about 10 cm or 4 inches. Does the Autopilot metadata provide a distance estimate for cars directly adjacent? If so, that would allow testing of Autopilot’s lateral accuracy as well.

Please keep in mind my request for this experiment! :) Some scientific results would truly inform the world’s perception of lidar and cameras!
 
In a lucky and unexpected reversal of fortune, Tesla decided to dissolve this latest blacklist for the unapproved v9 users. And to bring us in, they also pushed the 18.39.6 to us. This is a very welcome development and I am really happy Tesla made this latter choice and hopefully they'll avoid the blacklisting route in the future as well.

Also after some more playing with ulc in the past week I found it to be mostly a novelty gimmick - good to try a few times, but definitely not something I would be using often since the car is often doing lane changes that don't make sense in the grand scheme of things, stays in the slow lane too much or moves to the left too close to an exit and such. I am not going to try and forcefully enable ULC for the time being as the result, since I need to watch the car like a hawk anyway, it makes things somewhat more stressful. Let's see when Tesla feels it's actually ready.
 

ULC = unconfirmed lane change?

the car is often doing lane changes that don't make sense in the grand scheme of things, stays in the slow lane too much or moves to the left too close to an exit and such.

Interesting. Also disappointing it doesn’t work well yet. Funny how this is a problem of routing logic. Not something I ever thought of in an autonomous driving context before, except standard GPS roadmap navigation.

But intuitively, this seems like it should be solveable because it seems like a problem that is easy to build rules around. Like “make sure you are rightmost lane X metres or X travel time before the exit you need to take”.

Did you try Mad Max mode? Maybe it would get out of the slow lane more often.
 
Last edited:
ULC = uncommanded lane change

I did not try MadMax mode, but it's more about how aggressively it pushes out of it's current lane when it wants to get out and there's other traffic. I do not live in a traffic jam country so I don't need it.

Another annoying thing is (and I don't remember about reading it in the manual, though I might have missed it) when the car indicates it wants to change a lane and I think it's stupid, I don't know how to tell the car "no, don't do it" short of turning autosteer off.
 
Also after some more playing with ulc in the past week I found it to be mostly a novelty gimmick - good to try a few times, but definitely not something I would be using often since the car is often doing lane changes that don't make sense in the grand scheme of things, stays in the slow lane too much or moves to the left too close to an exit and such. I am not going to try and forcefully enable ULC for the time being as the result, since I need to watch the car like a hawk anyway, it makes things somewhat more stressful. Let's see when Tesla feels it's actually ready.


tenor.gif


Looks like i won't be kneeing to anyone, anytime soon.

I have never been wrong before. if you can show the car handling these things effortlessly that i listed above and what amon talked about. Then I will get on my knees and repent!
 
Anyway, those points in the clouds, poles and traffic signs... Seems to me Tesla gave up SLAM and went for something different

Those points on poles and signs could be for visual HD maps, which of course are used for visual localization. No idea why the clouds would have points on them. Maybe just training the system to filter out the sky?

I do not live in a traffic jam country so I don't need it.

Wait, are you testing it in the United States?
 
In a lucky and unexpected reversal of fortune, Tesla decided to dissolve this latest blacklist for the unapproved v9 users. And to bring us in, they also pushed the 18.39.6 to us. This is a very welcome development and I am really happy Tesla made this latter choice and hopefully they'll avoid the blacklisting route in the future as well.

Also after some more playing with ulc in the past week I found it to be mostly a novelty gimmick - good to try a few times, but definitely not something I would be using often since the car is often doing lane changes that don't make sense in the grand scheme of things, stays in the slow lane too much or moves to the left too close to an exit and such. I am not going to try and forcefully enable ULC for the time being as the result, since I need to watch the car like a hawk anyway, it makes things somewhat more stressful. Let's see when Tesla feels it's actually ready.

@verygreen Did you test ULC prior to the v9.0 release? If so, is it performing the same as before the black list or worse?

@Bladerskb Elon tweeted that Tesla is ironing out edge cases for Drive on Navigation / ULC. Since when did a tweet not work out well for Elon or Tesla? You may want retract that gif when Elon irons out these edge cases in the 2020 v10.0 release ;)
 
Awesome. Any way to pull the info for the side cameras and apply the info to what the side cameras are seeing like you’ve done for the front?
Yes. This is nearing completion now. I am trying to get some interesting/diverse footage. There are going to be some limitations of course, and the backup camera is so unreliable I think I am not going to even try to depend on collect that particular footage.