N1.09.4tracking loss fallbackdesignresearch

Loss of positional tracking needs a degraded mode, not an interruption

Aliases: tracking lost · 3DoF fallback · pose freeze · tracking degraded mode

What it is

Point the cameras at a blank wall or cover them with a palm and the position estimate dies first; angular rate usually remains. Halting the whole experience — black frame, freeze, eject to the system menu — removes space while the person is still standing and still turning their head. What is needed is a tracking-loss fallback: pin the last trusted position, keep updating orientation from the IMU, tell the wearer in a visible state that they can look but not walk, and rejoin when localisation returns.

An interruption treats the fault as an ending. A degraded mode treats it as a short, still-viewable, still-turnable regime.

Why it happens

In visual–inertial odometry, position is constrained by feature matches; orientation can coast on the gyro for a short interval. When texture vanishes, exposure fails, or a lens is covered, covariance blows up in translation first. If the runtime keeps moving the world from an untrusted position, the room slides or pops. If it freezes the whole pose, the world rides with the head during a turn — a worse fault than lost position.

Degradation stops the untrusted degrees of freedom and keeps the trusted ones. World-locked objects stay at their last coordinates; a head-locked status strip can still sit in view and say tracking is lost. On recovery, slamming in the new position makes the world jump a few centimetres, which reads as a collision. A short transition (fade, or spreading the correction over a fraction of a second) is required; do not teleport the visor the instant covariance drops.

A black frame is also a safety problem in standing use: the real room is still there and vision is not.

Studying it

Inject controllable localisation failures into a live session: cover the cameras, kill the room lights, or drop translational measurements in software for N frames. Compare blackout, full-pose freeze, and position-pin-only.

Independent variables: duration of loss, whether the head was turning or the body walking at onset, fallback policy, whether recovery is smoothed. Dependent variables: subjective break reports, world jumps during head motion, magnitude of the post-recovery snap, pause before the wearer resumes the task.

Safety observations belong in the same log: at the instant of loss, did the person reach for real furniture, did they doff immediately. Those tell you more about a bad interruption policy than a questionnaire.

Where it stops holding

A rotation-only device has no “position lost” path; the issue is signal drop itself. Occluded outside-in beacons and inside-out feature loss both kill position first, but recovery time differs, so the fallback needs a modality-specific cap. If loss hits while the wearer is walking with a virtual object in hand, pinning position desynchronises hand and object; freeze the grab or convert it into an explicit “grasp interrupted”. Lab occlusions under a second underestimate the multi-second blank-wall cases of real rooms.

Applying it

  • When position is untrusted: pin the last position, keep orientation, and stop world-locked content from sliding on a bad translation.
  • Show “tracking lost” as short copy or an edge tint, not an icon alone — the wearer is hunting for a spatial reference, and a readable state beats an abstract glyph.
  • Forbid a world teleport on recovery: corrections larger than a few centimetres fade or interpolate; a held object either comes along or is released first.
  • How to check: cover the tracking cameras for about three seconds, then uncover. The image should still follow head yaw, position should stay pinned, and a lost-tracking cue should be visible; after uncover, the world should not jump. A black frame, a full freeze, or a visor jerk on recovery all count as fallback failure.

Related

  • Same group: N1.09.1 Yaw drifts over time and needs a recentre entry point · N1.09.2 The origin definition decides which side of the person content appears on · N1.09.3 Panoramic footage is inherently 3DoF; translation cannot be added in post
  • Nearby: N1.02 Degrees of Freedom · N2.02 Bare-hand Tracking
  • Search terms: tracking loss · degraded mode · positional tracking

Cards in the same group

Quick Actions

Share

Share this page

ios_share

https://hci.top/en/handbook/N1.09.4