The commit action must not shift the gaze point
Aliases: confirmation gaze shift · looking at OK · commit-induced saccade
What it is
Gaze-plus-confirm only holds if, at the instant of commit, gaze is still on the indicated object. If the confirm action itself pulls the eyes away—to hunt an on-screen OK, to watch the hand, to line up a touch that must be looked at—pointing dies before commit. The system then selects the confirm control, or a neighbor the eyes were dragged onto, not the object just aimed at.
Why it happens
In eye–hand coordination, when the hand moves, the eyes often look first toward the hand or toward where the hand will land. An on-screen confirm button turns that reflex into a mandatory saccade: to press it you must look at it. On a trackpad, if the click zone is far from the current finger, the reach often takes attention and frequently gaze with it. In a headset, if pinching requires watching the hands to confirm the pose, the fovea leaves the object in the scene.
The commit channel therefore has a hidden constraint: its action loop must not presuppose “look somewhere else.” A physical key, a trigger near the body midline, a pinch that does not need visual alignment, a spoken phrase, satisfy it. An OK in the opposite corner, a gesture tutorial that must be stared at, and a confirm that is itself a second gaze target, violate it. After a violation the system samples the wrong coordinate at commit, and one saccade zeros the benefit of splitting point from commit.
Studying it
In a gaze-point plus confirm task, make confirmator type the independent variable: a key independent of gaze, an on-screen button that must be looked at, a gesture that must be watched. Align gaze coordinates at the commit instant with the original target, yielding an angular deviation at commit. Outcomes: that deviation, counts of selecting the confirm control by mistake, neighbor hits. Pfeuffer and colleagues already noted that touch actions disturb gaze in gaze-touch combinations; VR pinch studies have to ask whether pinching comes with looking down at the hands. A lab that places the confirm key where participants already know it by touch will underestimate the shift of the on-screen-button scheme.
Where it stops holding
Skilled touch typists can blind-press, and the shift from on-screen confirm drops—but that muscle memory cannot be assumed. People with low vision or a narrow headset field of view more often need to watch the hands to know a pinch happened, and the shift grows. Making confirm a dwell on the original target avoids the shift, but that is a retreat to dwell selection, not a combination. When spoken confirm fails in noise, people fall back to hunting an on-screen button, and the shift reappears as the fallback.
Applying it
- Put commit on an effector that does not need another look: a trigger already in the hands, space bar, a pinch that does not depend on watching the hands, a short phrase.
- Do not make OK an on-screen button that can only be hit by leaving the original target; if a second confirm is required, put it inside the original target or where gaze need not move.
- Verify by aligning commit timestamps with gaze samples; count the fraction that deviate by more than half a target width, and hits on the confirm control.