Regression: camera Y position readback incorrect when in VR

[machenmusik]

As mentioned in Slack...

mchen [5:53 PM]
I can't recall where exactly, but I think that current master has an issue where the camera position in VR is not getting the correct height value (missing the standing matrix offset?) when you read back its position via getAttribute('position') or .object3D.position?

dmarcos [5:55 PM]
@mchen those issues should be gone

mchen [6:02 PM]
https://codepen.io/machenmusik/pen/oEQbab
1. run Firefox and go to that codepen
2. enter VR and put on your Rift
3. look for the Y label, look down, and notice it's negative

dmarcos [6:03 PM]
It is the rendered height wrong?

mchen [6:03 PM]
the text should be headlocked, as a child of entity with active camera.
both the head sphere and the headlocked text are displayed below where they should be, which is why I had to look down
(the headsphere should be rendered where the HMD is, but since you're inside it, you may only see the nose cone)
the value appears to be off by roughly 1.6 :confused:
https://github.com/aframevr/aframe/pull/3327#issuecomment-360019236
The HMD height assertion in (1) is not correct, as demonstrated by this example.

mchen [6:25 PM]
this version shows controllers, for reference... https://codepen.io/machenmusik/pen/QQJyXV

• A-Frame Version: latest master
• Platform / Device: Windows 10 / Oculus Rift
• Reproducible Code Snippet or URL: https://codepen.io/machenmusik/pen/QQJyXV

[dmarcos]

Is not that the expected behavior? The height of the camera should apply automatically if you make the hud a child of it. If you add 1.6 then the hud will be at 1.6 + the height of the camera as you described.

[machenmusik]

from #3327 (comment)
The "floor" is essentially at y=100, so if my height is 1.51 meters in a 6DOF device the #cameraRig position will be left alone, but the #head position will change to 0 1.51 0, for a cumulative world position of 25 101.51 25.

What actually seems to be happening is that regardless of camera entity's position originally,
(1) the value gets overridden according to HMD (good)
(2) the Y value is negative, and appears to be off by roughly 1.6, so if I am sitting down and my HMD is about a meter off the floor, the Y position reads back (incorrectly) as between -0.5 and -0.6 -- per our discussion on that prior PR, I expect the Y position to read back as about 1

[machenmusik]

The hud text is positioned correctly with respect to the headsphere (good),
but the headsphere is in the floor (bad)

[dmarcos]

Ok... but going back to what you were trying to accomplish. If you want to head lock entities they should be just children of the camera. Is not that working as expected?

[machenmusik]

The headlocked hud text is not locked correctly to the camera, although it is a child of the element containing the camera component.
The headlocked hud text is positioned correctly relative to its headsphere sibling, but both of them are positioned well below the camera, as described earlier.

The scene does appear to be rendered with the camera at the correct height, and the controllers seem correct as well... but the entity with the camera component seems to be getting the wrong Y position value.

For convenience, the scene:

<a-scene>
  <a-sky color="#3cf"></a-sky>
  <a-entity id="player">
    
    <a-entity id="hmd" camera position="0 1.6 0">
      <a-text id="hud" position="0 -0.15 -0.25" wrap-count="80" align="center" value="Hi"></a-text> 
      <a-sphere opacity="0.5" scale="0.1 0.125 0.075">
         <a-cone opacity="0.5" color="magenta" position="0 -0.25 -1.05" scale="0.03 0.3 0.04" rotation="-90 0 0"></a-cone>
      </a-sphere>      
    </a-entity>

    <a-entity id="left-hand" laser-controls="hand:left">
     <a-text rotation="-90 0 0" wrap-count="80" align="center" value="L"></a-text> 
      <a-sphere opacity="0.5" scale="0.025 0.02 0.1">
        <a-cone opacity="0.5" color="red" position="0 -0.25 -1" scale="0.5 2 0.5" rotation="-90 0 0"></a-cone>
      </a-sphere>
    </a-entity>
    
    <a-entity id="right-hand" laser-controls="hand:right">
     <a-text rotation="-90 0 0" wrap-count="80" align="center" value="R"></a-text> 
      <a-sphere opacity="0.5" scale="0.025 0.02 0.1">
        <a-cone opacity="0.5" color="blue" position="0 -0.25 -1" scale="0.5 2 0.5" rotation="-90 0 0"></a-cone>
      </a-sphere>
    </a-entity>
    
  </a-entity>
</a-scene>

[machenmusik]

Let me know if these expectations are incorrect:
1 - the player rig position essentially defines the floor
2 - its child entity with camera component should receive the HMD's position, including standing offset from the floor (so if I told Oculus I am 1.5 meters, and the HMD Y position is currently reading 0.5 meters below that point, the child entity should have .getAttribute('position').y = .object3D.position.y = 1
3 - its child entities with controller components should receive the controller's positions, including standing offset from the floor
4- so the relative Y position of controllers and HMD should thus be correct.

At the moment, 1 and 3 are true, but 2 is not, so 4 is not.

[machenmusik]

the strange thing is, the camera rendering appears to be done from the correct Y position, but the entities that are supposed to be placed at the position of that entity with the camera are instead being placed in the floor. Which is why you find yourself looking down at the headsphere rendered beneath the camera vantage point.

[DougReeder]

Working with master last week, I observed an issue which appears to be the same. Specifically, on desktop (FF &Chrome under OS X) the camera position is correct, but in VR mode in Gear Vr (Oculus & Samsung) the camera appears slightly more than 1.5 m above where it should.

Runnable demo using A-Frame master at http://elfland-glider.surge.sh/arches/index.html

Source code using a "stable" version of A-Frame between 0.7 and 0.8: https://github.com/DougReeder/elfland-glider

There's an attitude indicator which should appear as triangle 1.5m above the user's head, outlined in black. On desktop and Android FF, it appears correctly. In this screenshot taken in VR mode in Gear VR Oculus, you can see the outline (or at least half of it) below the camera: https://hominidsoftware.com/non-public/bug-GearVR.jpg

(There's also a square of random color that appears where the avatar's waist should be.)

[machenmusik]

and stranger yet, today it is working as expected (camera and headsphere are always locked so you see the nose cone properly, headlocked text appears properly)

[dmarcos]

Just tested on Oculus and both cone and text are headlocked for me on https://codepen.io/machenmusik/pen/QQJyXV

[machenmusik]

ok, now I am seeing a weirder state - camera Y position shows as correct, and controllers are displayed at the right height, but the controller y positions are in the floor
https://gainful-pirate.glitch.me/

[machenmusik]

I modified the codepen to show HMD and controller Y positions, and I see the weirder state I was just describing...

[machenmusik]

#3407 may fix this?

[brianpeiris]

Yup, I believe #3407 should fix the controller y position.