Actions near the boundary should receive progressive cues rather than failing abruptly at the edge

What it is

As the hand approaches a face of the interaction volume, feedback should move continuously from “fully valid” toward “about to leave,” not jump from 1 to 0 on the face. Progressive cues let the user feel resistance or warning before crossing; abrupt failure is full function right up to the plane, then all commands gone one centimetre further. Mid-air has no physical lip; progressive cues are that lip built in another channel.

Why it happens

A hard boundary makes containment a boolean. A continuous drag is cut on the face; the value either freezes on the last frame or snaps to a default. Motor control needs an error signal before the irreversible point in order to reverse in time. A progressive band can encode distance with cursor opacity, falling control gain, weakening haptics, or a low “near the edge” tone. A band that is only 1 cm wide is as good as none; a band that is too wide puts a large interior region in “half-valid” and kills precision. What is being expressed is distance to the boundary, not recognition confidence—low confidence should take a different cue, or users will hear “I am about to leave” as “I am doing it wrong.”

Studying it

Have participants scrub along a direction normal to one face, comparing a hard cut with progressive bands of several widths. Measure whether they reverse before crossing, crossing count, continuity of the adjustment curve at the boundary, and “I could feel the edge” ratings. Build the progressive channel as visual-only, audio-only, haptic-only, and combined. Turn off or hold constant any visualization of recognition confidence so the two gradients do not tangle.

Where it stops holding

A discrete tap has no “still dragging near the boundary” phase; a progressive band helps taps less than continuous work, and taps need an account after crossing more than a gradient. When a body anchor sways with breathing, the band will flicker on its own and distance needs a temporal low-pass. Approaching two faces at once (up and right) requires the two gradients to compose, or the user will only repair one edge. Far-field setups with no display must put the gradient on a non-visual channel, or the requirement idles on that class of device.

Applying it

• Add a configurable near-boundary band for continuous operations, defaulted from hand tremor at that distance so ordinary shake does not chatter across a hard edge.
• Encode only geometric distance in the gradient; do not map classifier score onto the same opacity.
• Accept with an along-the-edge scrub: gain may fall and the cursor may fade; a jump from full effect to zero is not allowed. If a recording shows the command vanish in one frame with no prior change, treat it as a leftover hard cut.

Related

• Same group: C4.22.1 The boundary itself is invisible; users perceive it only through whether actions take effect · C4.22.2 Boundary crossing needs its own feedback, distinct from recognition-failure feedback · C4.22.3 Without boundary feedback, users misread out-of-bounds as a system fault or a wrong gesture
• Adjacent: C4.20 Interaction volume versus capture range · C4.26 Distinguishing three-state feedback
• Search: progressive cue · soft boundary · cliff edge