Add Scene Description API for automation and a11y

[AdaRoseCannon]

Right now rendered scenes are pretty opaque. They are hard to parse by machines to extract information about what is being shown and where it is in 3D space.

I would like to propose a solution where we have an object graph created by the user and attached to an entry point on the session each object is assigned a colour.
And a stencil buffer where these colours are rendered so that the device knows what is on the scene.

• Does this sound useful?
• Does it sound interesting?
• The spec side is pretty light but we should describe what is expected in the graph
• • What kind of information would you want, e.g. visibility, bounding-box
• • what kind of description is useful, we should make this extensible
• • Should the user pick the colours or should it be generated by a hash
• What carrots should we provide to get developers to actually use it?

/facetoface

[toji]

Mentioned in an editors meeting: There's a possibility that this information could also be used as a generic input assist, where we could start surfacing which semantic object a target ray intersected with select events. This could make some types of inputs easier for developers.