3D Selection Vault

Confidence-Aware Gaze-Depth Selection

Also known as Confidence-aware gaze-depth compensation

Eye gaze picks a target both by direction and by depth, estimated from where the two eyes converge; when the eye data look unreliable, the system widens its angular and depth tolerances, snaps to nearby depth layers and tells the user, instead of guessing.

How it works

On a Meta Quest Pro, a gaze depth is computed from the two eye rays, calibrated per user and filtered. A confidence score from fixation stability, head speed, target depth and contrast sets the windows: from 3.0° and 0.65 m at full confidence up to 6.8° and 1.55 m when confidence is low. When confidence is low and the scene is ambiguous, the depth estimate is blended with a local layer reference. Targets must fall inside both windows and be visible; the one with the best combined angle-and-depth score is the candidate, and if none qualifies nothing is selected. Status cues show low confidence, no candidate or waiting, and a dwell commits. With 60 participants (6,000 trials, targets at 0.5–6 m), it reached 19.1 successful selections per minute against 5.5 for a narrow fixed-window baseline, 12.2 for fixed wider windows and 13.0 for the same pipeline with a controller confirmation on every trial, which was the most accurate but slower and more demanding.

Papers