The WebXR Device API provides uniform access to 6 Degrees of Freedom (6-DOF: 3 axes of translation X, Y, Z, and 3 axes of rotation pitch, yaw, roll) via XRViewerPose and XRPose objects obtained on each XRFrame. The underlying runtime (e.g. OpenXR on Meta Quest / Apple Vision Pro, ARCore on Android, or ARKit on Safari) runs Simultaneous Localization and Mapping (SLAM) with visual-inertial odometry (VIO) to update the head pose relative to an XRReferenceSpace.

An XRReferenceSpace establishes the coordinate frame origin. In "local", the origin is centered near the viewer at session creation, ideal for seated VR experiences. In "local-floor", the origin is pinned to the physical floor directly below the user (y = 0 at ground level), essential for standing room-scale experiences. In "unbounded", the origin allows free roaming through buildings or outdoors without coordinate recentering, used in warehouse-scale AR.

As SLAM algorithms explore an environment, loop closure corrections cause previous coordinate estimates to adjust, leading to visual "drift" or sliding of virtual models placed in world coordinates. The WebXR Anchors API (frame.createAnchor) instructs the underlying native AR engine to continuously adjust the anchor pose to remain locked to physical real-world features and detected plane boundaries.

Each XRView in pose.views provides an eye property ("left" or "right"), a transform matrix offset by half the user's Interpupillary Distance (IPD / 2), and a projectionMatrix computed from the headset's asymmetric field-of-view (FOV: up, down, left, right angles). Applications render the 3D scene twice per frame (or use multiview extensions OVR_multiview2) into the XRWebGLLayer framebuffer.