02 · Environment
Training starts in familiar spaces.
The system uses kitchens, bathrooms, bedrooms, living rooms, and related daily environments rather than abstract visual-search tasks.
01 · Selected Research
LumaStep
Gaze-Mediated Virtual Reality Training for Visual Attention and Gaze Stability in Older Adults.
Research Question
Can visual training become a more explicit part of fall prevention?
LumaStep started from a simple observation: avoiding a fall depends not only on balance and gait, but also on noticing hazards, directing attention efficiently, and keeping vision stable while the head moves. I therefore built two training modules—environmental hazard search and gaze stabilization—and used eye tracking directly in the interaction.
03 · Hazard Search
Fixation becomes part of the decision.
Users inspect a scene and confirm a potential hazard through gaze. Dwell-based confirmation requires sustained attention rather than a quick pointer click.
04 · Feedback
Correct and incorrect choices are explained.
The task is instructional rather than purely score-based. Feedback identifies why an object may or may not contribute to fall risk and guides the user back into the search process.
05 · Variation
The scene changes before the interaction does.
Different room layouts and day/night conditions increase visual-search demand without forcing users to learn a new control scheme.
Training Architecture
One system, two visual demands.
Hazard search trains where attention goes in a scene; gaze stabilization trains whether vision remains stable during head movement. The two modules share the same intervention goal but operate at different levels of visual behavior.
06 · Gaze Stabilization
The second module isolates gaze stability from environmental search.
A table-tennis scenario provides a natural moving target, while a controlled fixation scene reduces environmental distraction. Together they train eye–head coordination under different levels of contextual complexity.
07 · Interaction Prototype
Try the dwell logic on the real training scene.
This web version simplifies the VR interaction: the pointer stands in for gaze and the radial timer stands in for dwell confirmation.
08 · Evaluation
Two studies separate interaction effects from training-structure effects.
Study 1 asks whether eye-controlled interaction adds anything beyond the same mixed training without gaze control. Study 2 asks how the two modules should be organized: mixed together, hazard search first, or gaze stabilization first.
Study 1Eye-controlled vs. non-eye-controlled
Study 23 groups × 3 time points
OutcomesBalance · gait · fear of falling
QualitativeVisual-strategy interviews
09 · Findings
The way the training is delivered matters.
Interaction mattered
Eye-controlled training showed greater improvement than the non-eye-controlled comparison on balance and fear-of-falling outcomes.
Training order mattered
Repeated-measures results indicated that the timing and ordering of the two modules influenced the trajectory of improvement.
Visual strategy changed
Interview data suggested a shift from relatively unsystematic scanning toward more goal-directed and spatially organized search strategies.
Research outputLumaStep: Gaze-Mediated Virtual Reality Training for Visual Attention and Gaze Stability in Older Adults — submitted to CHI.