Total eyes-off-road time is a separate constraint
Aliases: TEORT · cumulative glance time · glance frequency
What it is
Every single glance can stay under a cap and the total can still fail. Total eyes-off-road time sums every departure from the forward roadway inside a task window. It is a constraint of its own, not a restatement of single-glance duration. Twelve one-second glances never touch a single-glance ceiling, yet they slice continuous road sampling into thin pieces. This entry is only about that cumulative take. It is not about how long one look may last, and not about how a task should be chunked—that is task structure.
Why it happens
Driving needs the road model refreshed on a cycle. Each short glance punches a hole in that model; if the holes sit too close together, the model never finishes rebuilding and the person drives on a partial picture. Cumulative off-road time also raises the cost of coming back: heading is just reacquired when the next look at the interface pulls the eyes away again. Glance count has its own price. Each landing, road or screen, has to be aimed. The denser the shuttle, the shorter the unbroken looks that actually drive. “All short glances” is therefore not an automatic pass. Short-and-dense, or short-and-many, still starves the driving task.
Studying it
Inside a window from task start to task end, sum the duration of every off-road glance, and also report glance count and the spacing between glances. Visual-manual guidance lists total off-road time beside a single-glance cap because the two fail separately: some tasks produce a few long looks, others many short ones that add up. Visual occlusion’s accumulated “shutter open” time is another estimate of total visual demand.
Independent variables: number of steps, information per step, whether the protocol forces a return to the road between steps. Dependent variables: total eyes-off-road time, glance count, inter-glance interval, lane keeping, probe misses.
Where the window is cut changes the sum. Counting from menu open to destination confirm is not the same as splitting each sub-search. Define the task boundary first; do not pick a flattering slice afterwards. Simulator tasks are often completed in one go and never interrupted by real traffic, so totals can be both more concentrated and, in other cases, shorter than on the road. Both biases exist.
Where it stops holding
A navigation run may accumulate many brief looks at the next exit, each nested inside driving samples; that is not the same total as one continuous search-and-type episode. Highly practiced cluster checks are frequent and tiny, and their sum can still undercut one novice touchscreen action. Steps done after stopping should not enter an in-motion total. Reporting only the sum, with no single-glance distribution, can hide one dangerous long glance inside a pile of short ones. Independence is not substitution.
Applying it
- After every glance is already short, still count how many looks the task needs and what they add up to. A many-step flow that is short at each step still needs steps removed or deferred.
- Force a return to the road between steps. Do not use a continuous animation to deliver the next item the instant the last one ends.
- Verify by marking task start and end and summing every off-road glance in that window. If the total is still high, cut step count and whether a step truly needs a look—not another few hundred milliseconds off each glance.