In-motion tasks must be completable across short glances

What it is

An in-motion task has to be cut into fragments that each advance from one short look, not a stretch of staring until it is done. That is glance-interleaved completion: the structure itself lets the person shuttle between road and interface, and each look leaves a finished sub-action. The subject is how the task is sliced, not how long one glance may last and not how much off-road time the episode sums to—those are glance metrics. A flow can last a while in clock time if every step can be done and put down in a short look.

Why it happens

Driving periodically takes the eyes back. If the interface only makes sense after a whole screen has been read, there is no legal moment to return to the road, and the structure turns short glances into wasted glances. In-vehicle dialogue principles give pacing to the driver: the system waits; the person chooses when to look again. An interleaved task wears its state on the surface—what is selected, what remains—so a short glance can resume without a reread. A non-interleaved task hides meaning in continuous reading, and every return is a hunt for the cursor. Vehicle-dialogue standards of the ISO family ask for this driver-paced structure. They are not a request to squeeze total duration to a magic number.

Studying it

Visual occlusion is the matching paradigm: the interface is visible only in short “shutter open” bursts, and the question is whether the task still advances. With eye tracking, the question is whether glances land on subtask boundaries or in the middle of a sentence. Contrast two structures: a list that must be read as a stream, and a wizard that confirms one step and can be put down.

Independent variables: how much must be seen at once inside a step, whether a step is self-contained, whether the system proceeds or waits while the person looks at the road. Dependent variables: completion under occlusion, how much each open window advances, whether a half-step can be resumed.

Shutter timing in the lab is fixed; on the road, returns are timed by traffic. The two kinds of “short look” are not the same distribution. High completion shows that the structure can be sliced. It does not show that the visual demand of the slices, added up, is already acceptable.

Where it stops holding

Rear-camera video and live speed are driving information that must be watched continuously; do not wizard-cut them. A voice-first path barely depends on interleaved glances, so this constraint is weak—until the spoken result still needs an on-screen confirm, which must itself be interleaved. Setup done while parked, and one-shot pairing flows, are out of scope. Under high automation people can look longer; the structure should still be put-downable, because a takeover will cut continuous reading without notice.

Applying it

• Build in-motion flows so each step can be decided and stored in one look: pick one item, flip one switch, hear one sentence and decide whether to look.
• Ask one decision per view. Sequential short looks beat a comparison table that must be held in one stare.
• When the person looks at the road, the interface waits. Do not steal the next step with a carousel or a timed option.
• Verify by covering the display on a short cycle, physically or in software. Any step that cannot finish in the open window, and cannot resume when covered, does not yet belong in motion.

Related

• Within the group: K6.03.2 Uninterruptible long tasks do not belong while driving · K6.03.3 Interrupted tasks must keep their place
• Adjacent: K6.02 Eyes-off-road Time · K6.01 Types of Driver Distraction · A5.14 Interruption timing and breakpoints
• Search terms: glance-interleaved · paced dialogue · visual occlusion