Lazy follow lets body-lock keep up without shaking

What it is

If a body-locked panel is rigidly tied to torso or head every frame, gait bounce, breathing, and tracker noise all write into the UI and type shakes. If it does not follow at all, two steps leave the panel behind. Lazy follow puts a deadband and a low-pass on the follow: small motion is ignored, large motion is caught up smoothly, so the panel stays with the person without shaking on every tremor. It is a filter inside body-lock, not a switch to another lock, and not the visible move when the lock itself changes.

Why it happens

Torso and head trajectories contain gait frequencies, heartbeat-scale micro-motion, and estimator noise. Interface content rigidly bound 1:1 to those signals paints a corridor across the retina. Vestibular input knows the body is walking; vision sees “the panel is shaking”; the two fight. Lazy follow is usually split: position uses a deadband plus a spring–damper; yaw slowly aligns the panel to the person’s front; pitch almost never follows — a nod must not throw the menu up and down. Too small a deadband and shake remains; too large and after a turn the panel sits outside the view for a long time, read as lost. The time constant should miss the main gait band (about 1–2 Hz): strip the bounce, keep the one-shot yaw of “I turned around”.

Studying it

Have people read body-locked type while walking, standing, nodding, and turning, comparing rigid follow against several deadbands and time constants.

Independent variables: angular and linear deadband, follow time constant, whether pitch is followed, walking speed. Dependent variables: reading speed and accuracy, retinal stability of the panel (eye tracking or panel angular velocity), rated shake, time to retrieve after a turn.

Head-bob measurements from gait work and HUD-stabilisation work in VR are the same family. A seated test will not see gait bounce; the walkthrough must include real walking or a treadmill.

Where it stops holding

Seated, head against a rest, rigid follow is already stable and lazy follow looks like the panel is slacking. Content that must stay registered to a body site (a wrist menu on a watch pose) cannot use a large deadband or the affordance to the hand breaks. Head-lock is a different job: that layer is supposed to follow the view, and lazy follow makes a reticle lag — aiming must not use it. Jumps from tracking loss will be dragged into a false slide by lazy follow; they need a reset, not a deeper filter.

Applying it

• For body-locked reading panels: lazy-follow position and yaw with a deadband; do not follow pitch by default.
• Tune the time constant to strip gait bounce and still catch a single turn within one beat, then accept on the target headset with a walking-reading task.
• Provide an “align in front now” reset so filtering cannot leave the panel behind the back indefinitely.
• How to check: read a passage standing, walking, and after a fast turn. Type that jumps while walking means the deadband or filter is short; a panel that stays out of view after the turn means the deadband is wide.

Related

• Same group: N3.08.1 Different parts of one interface can use different locking modes · N3.08.2 A change of locking mode needs a visible transition · N3.08.4 Only world-locked content can be jointly pointed at by several people
• Nearby: N3.01 Locking Modes · N2.07 The Body as Reference Frame
• Search terms: lazy follow · body-locked smoothing · head bob