MR app to obscure exterior views from a moving simulator rig

My cousin implanted an itch I'd like to scratch. She does some academic psycho-neurological research that I don't fully understand. But in talking it sounds like mixed reality in the AVP might facilitate a type of research that she's pursuing. I just don't know if such a thing is currently possible with the AVP given how it treats physically moving frames of reference.

The idea is to have subjects in a motion vehicle simulator rig, able to operate it and see the interior of the vehicle/gauges/etc. But at times and in specific ways, obscure the views out of the "windows" of the vehicle. The obvious problem here is that the AVP doesn't like being in a dynamically moving frame of reference. This seems to be beyond what "travel mode" is intended for. They need to bump and turn the subjects to get the responses they're studying, so not a smooth ride like on a plane or train.

The interior of the simulator rig is a "known object" and can be modeled. I just watched the new video about training/tracking a hand-held object via Create ML.

Could a similar approach from a hand-held object be applied to the user's surroundings - the mocked up vehicle interior and its window frames?

We'd then apply an obscuring blur or even just an opaque polygon to that "window" while the simulator rig (and thus the user) is in motion? The alignment of the blur/polygon doesn't need to have perfect tracking and registration. Update rate does not need to be low millisecond, but full second updates might make motion sickness worse. Also, am I correct in inferring that object tracking should be free from major drift over the course of tens of minutes?

A stretch: would it be possible to counteract the illumination from room light sources so that the movement of light/dark is reduced on the interior of the "vehicle" while it is moving?

I also noticed in the new "Explore enhancements to visionOS object tracking" video that some of the effect of overlaying MR elements onto something that is visually passed through is being demonstrated on an iPhone. Could a proof of concept of this app(?) be "mocked up" using an iPhone as a basis to justify the expense of buying the AVP? Hold the phone while the rig is moving, track the interior and blur/"open" the windows as seen on the phone screen?

Quick takeaways from the docs and recent WWDC sessions:

  • Moving frame of reference is mostly a world-tracking problem, not an object-tracking one. Object tracking reports the cab's pose relative to the headset cameras every frame. If occluders are anchored to the tracked-cab entity (not world space) inside a fully immersive space, they should stay glued to the window frames regardless of rig motion.
  • Workflow: CAD-accurate USDZ of the cab → train a .referenceobject in Create ML (extended mode, all angles) → ObjectTrackingProvider in an ImmersiveSpace. For a moving rig, enable highFrameRateTrackingEnabled and tune movingObjectTrackingRate (requires the object-tracking-parameter-adjustment entitlement).
  • No long-session drift. Object tracking re-detects against the trained model each update — no integration, no monotonic drift. Practical limits are momentary loss-of-track from occlusion/glare, not accumulating error.
  • Room-light flicker on the interior: displays are additive, so real passthrough light can't be dimmed selectively. The clean path is to render an opaque virtual skin over the interior anchored to the tracked cab, decoupling its appearance from room lighting. Same approach for window occluders.
  • iPhone POC is viable. iOS 27 supports ARKit object tracking with the same .referenceobject assets. A phone-mounted prototype would validate tracking under your motion profile before committing to AVP. Caveat: monocular narrow-FOV passthrough — proves tracking/occlusion feasibility, not vestibular-visual equivalence.
  • Watch-outs: featureless interior surfaces track worse (may need fiducials); anything that physically moves (wheel, pedals) violates the rigid-object assumption and should be excluded from the reference model.

WWDC

Docs

MR app to obscure exterior views from a moving simulator rig
 
 
Q