Switching thresholds need hysteresis

What it is

If far mode and near mode share one distance line, a person rocking on that line—plus a little sensor noise—will flip the UI back and forth. Hysteresis splits the enter threshold from the exit threshold: closer in to enter near, farther out to leave it. The band in between keeps the current state. This entry is only the stability of the switch. It does not re-assign which input belongs at which distance, and it does not say how dense type should be after the switch.

Why it happens

A person is not a point on a ruler. Gait sways, waiting shifts a half step, talking to a companion turns a shoulder, and a depth camera’s range to the same body jumps. A single threshold treats that band of motion as repeated crossings, so the mode chatters: a menu opens and shuts, a touch layer flashes in and out, the attract loop and the operating UI fight for the same frame. Each flip is an attention capture and a cancellation of an act in progress.

Hysteresis is a Schmitt trigger in the control loop: two lines enclose a dead band, and the current state remembers how you entered. The cost is that one physical position can map to two UIs, depending on which side you came from. People can learn “a little closer and it becomes touchable.” They cannot learn “I did not move and it keeps changing.” Chatter hurts more than bistability because it cannot be attributed to one’s own motion.

Studying it

Have people walk naturally near the threshold, or do approach–stop–ease back–approach again, and compare a single threshold with a hysteretic pair. Engineering write-ups of proxemic systems treat flip count, and flips per approach, as usability incidents rather than as a precision problem in the sensor.

Independent variables: gap between enter and exit thresholds, ranging noise, lingering and lateral movement on the boundary. Dependent variables: mode flips per approach, operations broken by a flip, reports of “it changed by itself.”

Tape on a lab floor parks people exactly on the threshold and maximises chatter. In the field people often cross rather than linger, so chatter is rarer—but that does not licence dropping hysteresis. Waiting, taking a photo, and making way all stop on the boundary. A gap that is too wide feels like “I am already on the glass and it is still far”; report sluggishness separately from chatter.

Where it stops holding

A seated scene whose distance barely changes has nothing to switch, so hysteresis has no object. When only two modes exist and a button or a floor mat makes the switch, intent is already externalised and sensor jitter no longer owns the mode. Several people in different distance bands at once cannot each keep a private hysteretic state on one screen; a global mode either follows the nearest body or is undefined, and hysteresis does not solve “who is the reference.” Wildly noisy depth data needs filtering first; hysteresis is not a patch on broken ranging.

Applying it

• Give every adjacent pair of modes two lines: the distance to enter near is shorter than the distance to leave it; between them, keep the current state.
• Log mode flips during an approach. If range did not monotonically cross, it is chatter, not a real walk-in or walk-away.
• Preserve in-progress local state (scroll position, a selected object) across a mode change, so a false retreat inside the band does not wipe work and provoke more oscillation.
• Verify by asking someone to stand naturally for half a minute on the line you believe is the boundary (waiting, checking a phone, making way). If the UI switches more than a rare flicker, widen hysteresis. If they are on the glass and still far, tighten the exit line; do not delete hysteresis.

Related

• Within the group: K7.05.1 Distance selects which input is available · K7.05.3 Information density must follow viewing distance
• Adjacent: I4.01 Debounce and throttle · A10.02 Mode error · K7.02 Discoverability of Interactivity
• Search terms: hysteresis · proximity switching · mode oscillation · Schmitt trigger