The boundary itself is invisible; users perceive it only through whether actions take effect

What it is

There is no visible wall in the air. The boundary of the interaction volume is itself invisible; users cannot feel it like a bezel or see it like a screen edge. The only evidence for “am I still inside” is the after-the-fact result of whether an action took effect. Perception is indirect and lagged: act first, then infer which side of the box you stood on from presence or absence of a response.

Why it happens

Content on the display plane is visible from both inside and outside the volume; vision does not outline the input space. Proprioception can say which side of the body the hand is on, not where the software box’s six faces were drawn. People therefore use the action loop as a probe: reach, watch the UI, correct stance or amplitude. The probe’s resolution is limited by how discrete the action is—a tap yields only “this one counted / this one didn’t”; a scrub can feel an approximate edge. Lag also comes from recognition and feedback delay: when the response arrives late, the person has already moved, and the probe sample does not match the position of the moment. Without an active expression of the boundary, the whole map is fitted in the head from those sparse success/fail points.

Studying it

Draw no box and give no stance instructions. Ask participants to find the space that still works. Record search strategy (fore-aft, left-right, bigger motions), error between the fitted subjective boundary and the true geometry, and time to a stable working region. Compare outcome-only feedback, a continuous cursor, and a projected floor outline. This measures boundary perceptibility, not how large capture range is—the capture envelope can be measured beforehand and held fixed.

Where it stops holding

When AR draws a translucent box into the scene, or a mat or light carpet is on the floor, the boundary becomes an object and indirect perception is no longer the only channel. Far-field control that is only auditory or haptic has no shared view, so outcome feedback is weaker and the mental fit is harder. Skilled operators remember the patch of air in front of the device and the cost of indirect perception drops; a first-time visitor at a public screen has no such memory. Children’s success/fail explanations are less stable; estimating their perception from adult search strategies will bias.

Applying it

• Do not expect users to “feel the edge.” If the product needs people to stay inside, externalize the edge as a projection, a floor mark, or a valid cursor that travels with the hand.
• Log each success/fail with hand position at the time, to see where the user’s mental boundary has drifted, rather than reading only overall success rate.
• In usability tests, omit the box briefing first and see how many people can find a working region from indirect feedback in two minutes. If they cannot, the boundary still exists only in a config file.

Related

• Same group: 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 · C4.22.4 Actions near the boundary should receive progressive cues rather than failing abruptly at the edge
• Adjacent: C4.20 Interaction volume versus capture range · C4.10 Sensing volume and interaction distance
• Search: invisible boundary · indirect perception · probe by effect