Decoupling pointing from commit can solve both speed and false activation
Aliases: gaze-hand split · pointing versus click · multimodal split
What it is
When pointing and commit share one channel, speed and false activation lock each other: a sensitive channel is fast and false-positive-heavy; a conservative channel is clean and slow. Decoupling lets gaze take the sensitive, fast pointing job and a default-off modality take conservative commit, so the two curves no longer share one parameter. Dwell selection cannot make both true at once—it has one threshold. Look-to-select sets both to sensitive.
Why it happens
Pointing time in the Fitts sense grows with distance and shrinks with width; the eyes often cross a display faster than a hand because the movement is small and low-inertia. Commit time on a key or pinch is close to a fixed action time and does not grow with target distance. Giving pointing to the eye cheapens Fitts’s distance term; giving commit to the hand removes “whether to fire” from fixation duration. Mean selection time can approach “eye pointing plus one press,” while false positives approach those of the key rather than those of gaze.
Both at once requires: pointing error not so large that commit hits a neighbor; commit not so slow that it eats the eye’s advantage; users understanding two steps rather than waiting for a dwell clock that is not there. Experts learn “eyes first, hand second”; novices often move the hand to the target before looking, and the gain appears only after that strategy settles.
Studying it
On the same targets and users, compare three parameterizations: dwell only (sweep the threshold), mouse only, and gaze-point plus key. Plot each speed–error frontier and ask whether the combination beats both single-channel frontiers at once. Zhai and colleagues’ MAGIC experiments, and later VR gaze-plus-pinch evaluations, compare frontiers rather than a single number. Practice must be enough: first contact with the combination is often slower than a mouse. Split geometry into “large targets, neighbors almost unhittable” and “dense icons,” because on dense layouts decoupling hits pointing error before commit error.
Where it stops holding
When pointing error exceeds target spacing, decoupling solves Midas and not neighbor-hits; the speed gain is spent on corrections. If the confirmator is high-latency speech, commit dominates total time and decoupling contributes little to speed. People who cannot press or pinch stably are excluded by the split. Continuous aiming in games (hold to fire) is not a one-shot commit; the model becomes “gaze aims, the hand holds trigger state,” not a discrete click.
Applying it
- Choose the scheme from two curves, speed and error: if dwell must become very slow to reach acceptable false activation, switch to gaze pointing plus an independent commit.
- Make target spacing larger than typical gaze error, or time saved by decoupling will be spent undoing.
- Verify on the stable post-training segment that both completion time and neighbor-hit rate are not clearly worse than dwell-only and mouse-only.