[x3d-public] X3D VR

[Andreas Plesch]

Let's start with the RenderedTexture node since it is in actual use.

References first:

http://doc.x3dom.org/author/Texturing/RenderedTexture.html

http://realism.com/blog/3d-google-cardboard

http://doc.instantreality.org/documentation/nodetype/RenderedTexture/?filter=None

http://castle-engine.sourceforge.net/x3d_implementation_texturing_extensions.php#section_ext_rendered_texture

http://www.xj3d.org/extensions/render_texture.html

All follow the initial xj3d extension but only x3dom uses the node for
stereo rendering by adding stereoMode and interpupillaryDistance fields.
For a stereo pair it is necessary to use two RenderedTexture nodes, for
each of the eyes, with 'RIGHT_EYE' or 'LEFT_EYE' stereoMode values.

Using the interEye distance the node can then construct offsets from the
viewpoint position for rendering.

My extension adds 'RIGHT_VR' and 'LEFT_VR' stereoMode values which trigger
direct retrieval of the current left and right view and projection matrixes
from the VR display via the WebVR API. I currently multiply this view
matrix with the view matrix derived from the viewpoint's position and
orientation. I believe this behavior is more compatible with mouse
navigation which is also relative to the bound viewpoint. Mitch suggested
that viewpoint orientation values do not apply for GearVR . This is
consistent with leaving position and orientation values at their defaults,
eg. looking straight ahead.

The interpupillaryDistance field can be safely ignored since the VR display
provided matrixes already take this into account (after a user went through
a runtime calibration procedure). Similarly, the fov and width/height
aspect ratios are device specific and already properly accounted for. In
addition, apparently there is some fine tuning of the projection matrix to
lens characteristics.

There is a failure mode when 'RIGHT_VR' is requested but no VR display is
available. Currently, I plan to fall back to 'RIGHT_EYE' .

There is another additional, SFInt field 'vrDisplay' in my extension which
defines the index of the webVR display to be used in case multiple devices
are available.

The RenderedTexture approach is a low-level solution which requires some
changes to a scene for VR purposes. It turns out it is possible to build
helpers in the form of injectors or likely as protos on top which make it
very easy to use this approach for existing scenes.

In terms of parallel rendering of stereo pairs the RenderedTexture approach
may not be ideal since the nodes for each eye are independent from each
other. Presumably, a smart browser could search for matched pairs of nodes
and combine their rendering but that does not sound very promising.

Andreas

On Dec 21, 2016 6:00 AM, "Roy Walmsley" <roy.walmsley at ntlworld.com> wrote:

> Hi Andreas,
>
>
>
> That’s great to see what you have been doing here. Would  you like to take
> your node suggestions one step further and propose the fields that might be
> specified for each node?
>
>
>
> What we could do then is, in the new year, is to host an open discussion
> on WebVR and X3D. Contributions from any other readers are welcome, and
> could lead to a lively discussion on this topic.
>
>
>
> Thank you for the suggestions,
>
>
>
> All the best,
>
>
>
> Roy
>
>
>
> *From:* x3d-public [mailto:x3d-public-bounces at web3d.org] *On Behalf Of *Andreas
> Plesch
> *Sent:* 21 December 2016 05:51
> *To:* X3D Graphics public mailing list <x3d-public at web3d.org>
> *Subject:* [x3d-public] X3D VR
>
>
>
> Hi
>
> Working with x3dom and WebVR I am exploring what additions to X3D would be
> necessary or useful to effectively use current VR hardware such as the
> Rift, Vive, Gear, or Cardboard.
>
> The proposed RenderedTexture node is currently used in x3dom with special
> stereo modes to generate left and right views from a single viewpoint in a
> single GL context.
>
> A Layer for each left and right view could also be used with two
> coordinated but separate viewpoints. This would be Cobweb's pattern.
>
> Since the HMD and its own runtime know best the internal optical
> characteristics of the display, the preferred interface at least in WebVR
> are view and projection matrixes directly usable by consumers for 3d
> graphics generation, and not position, orientation and fov. This leads to a
> new requirement for X3D to accept these raw matrixes as input.
>
> Since x3dom uses RenderedTexture for VR, I added there additional special
> stereo modes which directly receive these matrixes from the HMD to be used
> for rendering at each frame. This in effect accomplishes a first, head and
> body based, level of navigation. In my first tests (on github) it works
> well and should be robust across devices. This approach does require some
> standard API to the HMD such as WebVR.
>
> Another possibility is to add view and projection matrix fields input
> fields to viewpoints which can be continually updated. One could convolve
> the view matrix with position/orientation fields, or optionally completely
> ignore them.
>
> One then could use external SAI to keep updating or perhaps introduce an
> environment ProjectionSensor. It would relay these matrix events from an
> HMD runtime to then be routed to the viewpoints.
>
> A second level of navigation is accomplished with handheld controllers.
> Until some standard gestures evolve, it will be necessary to expect custom
> per scene navigation. X3d should have a way to sense buttons and axes
> velocities of such controllers so these are then available to manipulate
> the view in some way. InstantReality has a general IOSensor to that effect.
> On the web the GamePad API is a standard which would need to be used.
>
> Summary: RenderedTexture with stereo modes, matrix fields for viewpoints,
> a ProjectionSensor, and a ControllerSensor would all be candidates for x3d
> VR support.
>
> Thanks for reading, any thoughts welcome,
>
> Andreas
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://web3d.org/pipermail/x3d-public_web3d.org/attachments/20161222/6326bb71/attachment.html>