[Proposal] Marker point-instance in gpu vs non-point-instance implementation

[ooctipus]

Proposal

Current IsaacRTX Marker is constructed using pointinstancer, whose transform and scale is updated through a CPU path

Point Instance assets has some special API that populate the scene with arbitrary number of instances at every call.
you can expect something like this to work :

given stage

/World/env/envs_0/.....
/World/env/envs_1/.....
/World/env/envs_2/.....
/World/Visualizers/Sphere

PointInstancer Instance

point_marker = UsdGeom.PointInstancer.Define(stage, "/World/Visualizers/Sphere")Visualization using 

PointInstancer in lab implementation will be:

def visualize(translation, orientation):
  point_marker.GetPositionsAttr().Set(Vt.Vec3fArray.FromNumpy(translation))
  point_marker.GetOrientationAttr().Set(Vt.QuathArray.FromNumpy(orientation))

you may see that this will make things very slow in high count of marker.

The ask is either

1. figure out a way to interact with point-instance in gpu native path without numpy getting our way for point instance.

or

2. if there is more faster way to visualize marker, we can switch to a new marker implementation that uses non-point-instance.

[RandomOakForest]

All good points, this has been duly noted.