Per-eye vergence data not available through XR_ANDROID_eye_tracking (combined gaze only)

[blondarb]

Summary

When using eye tracking via the XR_ANDROID_eye_tracking OpenXR extension on Samsung Galaxy XR, the left and right eye gaze data appear to be identical (combined/averaged gaze duplicated to both eyes). We need independent per-eye rotation data to measure binocular convergence (near point of convergence).

Use Case

We are building a clinical neurologic exam app (SDNE) that includes a fixation/convergence assessment task. Convergence measurement requires computing the vergence angle between the two eyes as a target approaches the patient. Without independent left/right eye pose data, this measurement is impossible.

What We Observe

• EyeTracking.LeftEyeRotation and EyeTracking.RightEyeRotation return values that appear identical or nearly identical
• Combined gaze direction works well for saccades, pursuit, and fixation tasks
• Confidence is reported as 0.95 (hardcoded?)
• Device: Samsung Galaxy XR (SM_I610)
• Unity 6 (6000.3.6f1) with OpenXR

Questions

1. Does the XR_ANDROID_eye_tracking extension support per-eye vergence data on Galaxy XR?
2. If so, is there a specific permission, API path, or extension that must be enabled to get independent left/right eye poses rather than combined gaze?
3. If not currently supported, is this planned for a future SDK release?

Environment

• Device: Samsung Galaxy XR (SM_I610)
• OS: Android XR
• Unity: 6000.3.6f1
• OpenXR runtime
• Permissions granted: EYE_TRACKING_FINE, EYE_TRACKING_COARSE

[blondarb]

This is the same underlying need as #15. We're building a clinical neurologic exam (SDNE) on Samsung Galaxy XR that requires per-eye gaze vectors for convergence/vergence measurement (cranial nerve III/IV/VI assessment).

@LevanaChen's response on #15 mentioned that fine per-eye poses through the Google Extensions package are in progress. We're very interested in this — convergence testing is currently disabled in our exam because the combined gaze from EyeGazeInteraction doesn't support it.

In the meantime, we'll investigate whether the coarse per-eye data from ARFace.leftEye/rightEye provides enough angular resolution for clinical convergence detection.

We've also started using TryGetAndroidOpenXREyeTrackingStates() for blink detection based on the information in this thread — very helpful, thank you.