TL;DR:
- Eye tracking enables dynamic foveated rendering — rendering at full resolution only where you’re looking — which can reduce GPU load by 30–50% while maintaining perceived visual quality
- Apple Vision Pro, Meta Quest Pro, and PSVR 2 all use eye tracking for both rendering optimization and gaze-based UI interaction; Meta Quest 3S introduced partial eye tracking support in 2026
- Gaze data raises significant privacy considerations — it’s biometric data that reveals attention, emotional state, and health indicators, requiring careful handling in enterprise deployments
The human eye doesn’t see uniformly. The fovea — the small central region of the retina — provides sharp, detailed vision in about a 5° arc of your visual field. Peripheral vision is progressively lower resolution. Your visual system compensates through rapid eye movements (saccades) that continuously repoint the fovea at things you’re interested in, and your brain constructs the experience of uniform, detailed vision.
This biology creates an opportunity for VR rendering. If a headset knows exactly where you’re looking — in real-time, with sub-degree accuracy — it only needs to render at full resolution in that small region. The rest of the field of view can be rendered at lower resolution, invisible under normal conditions because your peripheral vision can’t resolve it anyway. The GPU work saved by rendering 90% of the frame at lower resolution is substantial.
This is foveated rendering, and eye tracking is what makes it work properly.
How Eye Tracking Works in XR Headsets
Modern XR eye tracking uses near-infrared (NIR) illumination and cameras focused on the eyes from inside the headset. The cameras capture the corneal reflection patterns of the NIR illuminators along with the pupil position and shape. Computer vision algorithms calculate gaze direction from these measurements.
Gaze accuracy: Consumer headsets typically achieve 1–2° of accuracy with calibration. Research-grade eye trackers achieve sub-0.5°. For foveated rendering, 1–2° is sufficient — the full-resolution region just needs to be a bit larger than the fovea to cover calibration error. For gaze-based UI interaction (using your eyes to select elements), 1–2° is enough to target reasonably sized UI elements comfortably.
Calibration: Most headsets require a brief calibration procedure when you first put on the headset, or periodically. You look at a series of targets while the system maps the relationship between raw sensor data and actual gaze direction for your specific eye geometry. Some headsets now use continuous implicit calibration that improves accuracy over the session without explicit user action.
Update rate: Eye tracking typically runs at 90–120Hz (matching the headset frame rate), fast enough to track saccades (which take 30–120ms) and smooth pursuit eye movements.
Foveated Rendering: The Technical Payoff
Without foveated rendering, a VR headset renders every pixel of a wide field of view at full resolution, every frame. At 2K×2K per eye at 90fps, this is enormous — the primary reason VR has historically required powerful GPUs.
Fixed foveated rendering (available even without eye tracking) renders the outer edges of the frame at reduced resolution based on where peripheral vision is. It’s a cruder approximation — the center of the frame is assumed to be where you’re looking.
Dynamic foveated rendering (requires eye tracking) shifts the full-resolution region in real-time to follow your actual gaze. Because you’re never looking at the outer edges, this region can be rendered at 2–4× lower linear resolution, saving 75–90% of the fragment shader work for that area.
In practice, DFR typically saves 30–50% of total GPU rendering work while being imperceptible under normal viewing conditions. This translates directly to either better visual quality at the same GPU budget, or lower GPU power consumption (important for standalone headsets running on battery).
Current Headsets with Eye Tracking
Apple Vision Pro: Eye tracking is central to the entire UI paradigm. “Look, pinch” is the primary interaction method — you look at an element to focus it and pinch your fingers to activate it. The system uses eye tracking for rendering optimization, UI targeting, and the privacy-preserving gaze model where even Apple’s own apps don’t receive raw gaze coordinates (they only know when an element is selected, not that you looked at it).
Meta Quest Pro: Eye tracking with dynamic foveated rendering support, plus social eye tracking — avatars in social VR environments have eyes that actually reflect where you’re looking. Good accuracy for enterprise applications.
PlayStation VR2: Eye tracking with adaptive foveated rendering for PlayStation 5. Sony’s implementation focuses on rendering efficiency rather than gaze-based UI interaction.
HTC Vive Pro Eye / Focus Vision: Enterprise-focused headsets with high-precision eye tracking, good for gaze analytics in training and simulation applications.
Samsung Galaxy XR (Android XR): Includes eye tracking with support for Android XR’s gaze-based accessibility APIs and foveated rendering.
Meta Quest 3S: Introduced limited eye tracking (tracking-grade, not gaze-based UI) in late 2025/2026, primarily for social avatars and fixed foveated rendering assistance rather than full dynamic foveated rendering.
Gaze-Based Interaction: Beyond Just Rendering
Eye tracking enables input modalities that didn’t exist in traditional computing:
Dwell-based selection: Look at an element for a set time (typically 0.5–1.5 seconds) to trigger it. Useful for accessibility and hands-free interaction.
Gaze + gesture: The Vision Pro model — look to target, gesture to confirm. Separates the “what” from the “do it” to reduce accidental activation.
Contextual attention: Applications can respond differently to content you’re actively looking at versus content in your peripheral vision. A detailed panel might show a preview when you glance at it and expand when you look at it deliberately.
Gaze analytics: In training and simulation, knowing where trainees look during simulated procedures is directly useful — surgeons looking at the wrong instrument during a simulated emergency, pilots not checking a critical gauge, etc.
Reading and document work: Apple Vision Pro’s visionOS handles text rendering with eye tracking data to optimise rendering quality for text you’re actively reading.
Privacy Considerations
Gaze data is biometric. Where you look, for how long, and in what pattern reveals:
- Attention and interest: Advertisers have long understood that eye tracking data is premium data for understanding what people actually pay attention to versus what they ignore
- Emotional state: Pupil dilation correlates with arousal, cognitive load, and emotional response
- Health indicators: Gaze patterns are used in neurological assessment (diagnosing concussion, Parkinson’s, certain cognitive conditions)
- Biometric identification: Gaze patterns are individual enough to serve as a biometric identifier
For consumer applications, the question is how gaze data flows. Apple’s approach — the OS handles gaze and apps receive only confirmed selections, never raw gaze coordinates — is meaningfully privacy-protective. Meta’s approach gives developers access to aggregate attention metrics but not precise gaze.
For enterprise deployments, organisations need to treat gaze data as biometric data under GDPR and similar regulations, requiring informed consent and appropriate data minimisation.
The Developer Perspective
For developers building for eye tracking-capable headsets:
visionOS (Apple Vision Pro): SwiftUI views automatically get eye focus states. The system manages targeting — you declare that a view is interactive, the system handles the eye tracking targeting and selection.
Meta SDK: Quest Pro exposes eye gaze data via the EyeTrackingBehavior component in Unity’s Meta XR SDK. Foveated rendering is enabled via OVRManager with a couple of configuration calls.
OpenXR extension XR_EXT_eye_gaze_interaction: The standard OpenXR extension for gaze data across runtimes. Increasingly well-supported across headsets that implement OpenXR.
Performance guidance: When implementing foveated rendering, test with the high-resolution foveal region that’s 2–3× larger than the stated accuracy — this provides a comfortable margin so the sharp region doesn’t visibly lag behind eye movement.
Eye tracking is transitioning from premium feature to table stakes for high-end XR hardware. The rendering efficiency gains alone make it a compelling addition, and the interaction paradigms it enables — particularly Apple’s look-and-pinch model — are proving genuinely more natural than controller-based pointing for many tasks. Expect it to appear in the next generation of Quest and Samsung headsets at consumer price points.