I wonder if a separate camera for pancake with a larger FOV would help, with the translation (e: and rotation) matrix lowpass-filtered from user orientation.
But I feel its mostly rotational motion that is perceptible in VR demos rather than translational motion. I suspect that's because a small rotational motion results in a large shift in perspective for objects in the world that are some distance away from the camera while small translational movements don't really result in a significant perspective shift.
Would be interested in seeing what research into human head motion while sitting/standing/walking/etc. shows though.
