With vision occluded, estimates of one's own position drift

What it is

Stand for half a minute in an occluding headset and the person still believes they are in the middle of the room; they have drifted to the TV cabinet. Without visual landmarks, spatial updating path-integrates from vestibular and proprioceptive cues, and error accumulates. The estimate of “where I am, and what my hand will hit” walks off on its own — before any swing, body coordinates and room coordinates already disagree.

Why it happens

With eyes open, walls, furniture, and light patches keep pinning self-position to room coordinates. Occlude those anchors and updating switches to vestibular integration plus the proprioception of steps and turns. Integration is biased: a short step over- or under-counted, a turn over- or under-estimated, and after a few minutes the self is somewhere else. If the virtual floor moves while the person is standing still, vision writes a displacement that never happened into the same update, and drift accelerates.

The hand drifts too. Unseen, the fingertip's estimate in body coordinates slowly diverges, so a reach toward a virtual cup points at a different place in the room. That is not the same fact as swinging into a coffee table that was always on the arc and simply unseen. Here the person still thinks they are on the safe spot, and the coordinates have left. Guardian alarms on tracked true pose; the person stands on estimated pose. When those disagree, the alarm feels unearned, or arrives closer than expected.

Studying it

Classic tests are blind walking back to origin and triangle completion: two legs, then point home, and measure endpoint error. In the headset, compare full occlusion, remaining real landmarks, and virtual landmarks aligned with the room. Standing still while the virtual world walks is a second condition, to split vestibular integration from visually written false displacement.

Independent variables: occlusion duration, real landmarks visible, whether virtual optic flow implies motion, whether stepping is allowed. Dependent variables: position error, heading error, error when pointing at a real corner, endpoint error when reaching toward a fixed real object.

Error grows with time. Report it by the minute, not as a single offset after one teleport.

Where it stops holding

Video or optical see-through keeps recalibrating on real landmarks, so drift is much slower. If the person barely translates and only turns the head, the translational term of path integration is small and what remains is slow heading drift. Some people pin themselves with nonvisual anchors (a steady fan, the HVAC), and individual differences are large. Tracker origin drift (lost environment understanding) stacks on the person's estimate — keep “where I think I am” and “where the system thinks I am” apart in the log.

Applying it

• In occluding sessions, give a room-aligned real referent every so often (a flash of see-through, real floor texture at the feet, a wall calibration). Do not let path integration run the whole session alone.
• When people must stand near real furniture, limit travel from tracked true pose, not from their sense that they are still centred.
• Do not let the virtual floor slide under a standing person; false displacement writes into spatial updating.
• How to check: after minutes of occlusion, have them point at a real corner or walk back to start. If error is already an arm's length, self-position has left the safe spot — add a calibration or shorten the landmark-free stretch, rather than painting the guardian brighter.

Related

• Same group: N2.13.1 Large gestures hit furniture and people the wearer cannot see · N2.13.3 Bystanders cannot predict the wearer's next movement · N2.13.4 Banning dangerous actions in the content is more effective than after-the-fact warnings
• Nearby: N4.04 Room Scale and Safety Boundaries · N5.06 Bystanders
• Search terms: spatial updating · path integration · proprioceptive drift