GazeIntent

Also known as GazeIntent-General, GazeIntent-Personal, Intent-adaptive dwell

Gaze-only dwell selection whose dwell time shrinks when a machine-learning model, reading your eye movements, predicts that you are about to select.

How it works

An LSTM intent model reads 17-sample windows of gaze velocity, displacement and acceleration (plus fixation and saccade flags) and predicts whether the user intends to select. The last four predictions are combined into a scaling factor that cuts the static dwell threshold, down to instant selection when the model is confident, so false positives from single predictions are damped. The model (F1 = 0.94) was trained on controller selections in an arithmetic task on a Meta Quest Pro, then deployed for gaze-only selection in a Fitts-style task, arithmetic grids and a sliding puzzle. Compared with static dwell and a dwell-then-intent method (after Isomoto et al.), GazeIntent allowed many more selections in the faster tasks, felt fastest, and personalised models were preferred by most returning users.

Papers

Anish S. Narkar, Jan J. Michalak, Candace E. Peacock, Brendan David-John (2024). GazeIntent: Adapting Dwell-Time Selection in VR Interaction with Real-Time Intent Modeling. Proceedings of the ACM on Human-Computer Interaction 8(ETRA), Article 226 doi:10.1145/3655600