Mid-air gestures are a poor channel for precise numbers and long text

What it is

Mid-air gestures fit a small set of discrete commands. They do not fit treating a finger as a dial for a precise number, and they fit long text even worse. With no surface to supply friction, detents, or tactile quantization, a fingertip jitters in air and the value the system sees hunts around the target. Letter-level text needs dozens of ordered discrete choices; segmentation error and fatigue compound on every character. Hanging volume, temperature, a search box, and a password on waves asks the channel for a resolution it does not have.

Why it happens

The arm is an open kinematic chain. Tiny tremor at shoulder, elbow, and wrist is amplified at the fingertip. Without a desk or screen to constrain motion to a plane, holding a pixel-level position takes continuous co-contraction. Numeric input also needs a mapping: how far the hand moves versus how much the value changes. High gain turns jitter into jumps; low gain makes a long path and a longer unsupported hold. Text is worse. Each character is a selection, and an in-air keyboard has no key-cap edges, so neighboring letter hit regions overlap in space. Speech or a physical keyboard hands granularity to another channel. Forcing mid-air to carry it makes error rate and time grow with string length rather than stay constant.

Studying it

Standard pointing tasks can measure effective target width in air, but precise numbers and long text need real tasks. For numbers, drag an in-air slider to a specified reading (21°C, 47%), and record final deviation, overshoot count, and time-to-stable. For text, copy a short sentence and a longer passage; record correct characters per minute, deletions, and mid-task arm drops. Controls are a touch slider, a knob, and a physical keyboard. Independent variables include gain, snap-to-tick, and whether a surface supports the arm. Do not claim the channel fits arbitrary input from a “pick three large icons” task.

Where it stops holding

With snap-to-tick or quantized steps, setting a coarse level (mute / low / high) is a discrete choice, not a precise number, and can work. In VR, using the other hand as a temporary surface or handing the value to a knob on a controller is no longer pure continuous mid-air input. Short PINs can occasionally succeed in good light; the cost of a miss (locked account, wrong room) usually beats the gain of skipping a keypad. Eye or head micro-adjustment can add precision, but that is already multimodal and must not be scored as the gesture itself becoming accurate.

Applying it

• Take continuous numbers and text off the default mid-air job list; hang them on knobs, sliders, speech, or a real keyboard.
• If a parameter must be set in air, use a small number of snapped ticks and show the current tick, not a smooth cursor. Do not ask an arm to paint an arc onto a ones digit.
• Accept with specified-reading and copy-typing tasks. If final deviation, overshoot, correct characters per minute, or mid-task arm drops fail, keep that function off the gesture path.

Related

• Same group: C4.01.1 Without contact, input is segmented from continuous motion · C4.01.2 Fit: contact is undesirable, distant control, or spatial objects · C4.01.3 Cost vs non-gesture paths must include discovery, entry, recovery, and exit
• Adjacent: C4.12 Fatigue cost of mid-air gestures · C6.01 Keyboard layouts
• Search: mid-air pointing precision · mid-air text entry · unsupported arm tremor