QuestNav: The Best Robot Pose Tracking System in FRC

[ultraprecision]

I am getting the Web Socket error I’ve followed every step as list on GitHub even tried on 2 different computers with fresh installs of all programs still get the same error.

Could someone please help ?

[Crafter]

Okay so after trying it on a Q2 it crashed after loading for a bit. The logs below appear to be relevant. @juchong, any idea what’s up?

Edit: It runs fine on the Q2 if I use “build and run” in Unity. However this is not a permanent solution.

[kn4]

I ran into this same issue. The project has the websocketsharp dll in a Git LFS file, so you will need to make sure you are properly using Git LFS upon cloning the repository.

[ultraprecision]

Thanks so much i got it to work now !

[iz_en_zi]

@AngleSideAngle

The conditions the quest were tested on were not as hostile as an actual FRC match (which likely contains impacts that the quest was not designed for), so it is to be seen how well the quest’s vision-based slam holds up.

Well, the quest 2 did survive this person’s dryer: https://m.youtube.com/watch?v=SwCS85BW8NQ&t=376s&pp=ygUYcXVlc3QgMiBkdXJhYmlsaXR5IHRlc3Qg

[Nanastasoff]

Does anyone have an example of reseting the pose on the quest to a location? Iv’e been having a hard time getting that working. Thanks!

[Chadfrom7226]

Can’t wait for the thread “TelsaNav: The Best Robot Pose Tracking System in FRC”

Followed by a picture of a robot full of Telsa cameras and GPUs

[Wihy]

Would still cost less than a limelight

[flametek]

And probably have better documentation for it too.

[juchong]

It looks like a few questions have piled up, so let me reply to them in a single post:

@Crafter

Would this work with a Quest 2?

Probably not. The recommended headset is still a Quest 3S.

@JohnFogarty

Is anyone working on incorporating the AprilTags into the algorithm?

I’m working on it, but life has gotten in the way a bit.

@D_Price

In the future, will you venture towards prototyping your own visual device similiar to this to possibly bring the cost down and compact the footprint a little on the robot?

No.

@jdaming

I believe you will currently (until meta gives direct access to the camera which I heard a rumor is supposed to be “early this year”) need to use this method

Correct. This is not a rumor, though. App access to pass-through cameras was announced at Connect 2024.

[brainiac]

Correct me if I am wrong, but wouldn’t April tag support be unnecessary if during initial match setup you place the robot near a reference field element, and then just manually set the reference pose to be that set location?

[Crafter]

@juchong

Probably not. The recommended headset is still a Quest 3S.

I actually managed to get it working. I’m not sure if anything is broken but it seemingly worked.

[JohnFogarty]

I admit I do not understand a lot about the underlying mapping software and technology in the headset but I assumed, possibly wrongly, that it could perform better if it did use the AprilTags in some way. The dynamic elements of the field like the swinging cages and other robots are also a factor that no one has tested here yet.

[juchong]

@brainiac

Correct me if I am wrong, but wouldn’t April tag support be unnecessary if during initial match setup you place the robot near a reference field element, and then just manually set the reference pose to be that set location?

You are correct. April tags would only serve to initialize the headset’s position on the field. The headset’s tracking pipeline would take over from there.

@Crafter

I actually managed to get it working. I’m not sure if anything is broken but it seemingly worked.

Oh, cool! The project currently compiles for a Quest 2, but I hadn’t ever tested it. Did you have to change any of the build/project settings?

@JohnFogarty

I admit I do not understand a lot about the underlying mapping software and technology in the headset but I assumed, possibly wrongly, that it could perform better if it did use the AprilTags in some way. The dynamic elements of the field like the swinging cages and other robots are also a factor that no one has tested here yet.

April tags might help with tracking if their measured positions were injected into the pipeline itself, but it’s unlikely that they’d improve performance when used at the application level. Dynamic field elements aren’t a problem for VIO-based SLAM.