Gaze-driven interaction — foveated rendering, attention-aware interfaces, gaze-based selection — only works as well as the eye tracker underneath it, and headset makers rarely publish independently verified error figures. A new study puts three devices through the same protocol: Meta Quest Pro, HTC VIVE Pro Eye, and the HTC VIVE Focus Vision — the last of which, the authors note, had never been formally evaluated before.
Thirty participants ran the same random-saccade task on each headset. Median mean error came out to 1.04° on the Quest Pro, 1.50° on the VIVE Pro Eye, and 2.41° on the Focus Vision — and the Focus Vision remained the least accurate even after linear recalibration, the step that normally closes most of the gap between devices. A second block with a darker background pushed VIVE Pro Eye error up by another 0.70°, suggesting ambient lighting conditions matter more for some trackers than others.
The more interesting finding isn’t the headline accuracy ranking — it’s what a simple error average hides. The Focus Vision’s samples were positively correlated (dispersion 9.8x the sample-to-sample root-mean-square), the Quest Pro paused for up to 6.04 seconds at a stretch, and during instructed eye closures the Quest Pro flagged every sample as valid even while the reported gaze position moved 9.74° — meaning a developer trusting the device’s own validity flag would have shipped bad data without ever knowing it. The authors’ recommendation: any eye-tracking evaluation should report repeated timestamps, loss composition on the device’s own clock, and validity behavior during eye closures — not accuracy alone.
This is an unreviewed preprint, not yet peer-reviewed; treat the specific figures as early but methodologically careful, not final.