WebXR Device API

1. Introduction

1.1. Terminology

This document uses the acronym XR throughout to refer to the spectrum of hardware, applications, and techniques used for Virtual Reality, Augmented Reality, and other related technologies. Examples include, but are not limited to:

• Head-mounted displays, whether they are opaque, transparent, or utilize video passthrough

• Mobile devices with positional tracking

• Fixed displays with head tracking capabilities

The important commonality between them being that they offer some degree of spatial tracking with which to simulate a view of virtual content.

Terms like "XR device", "XR application", etc. are generally understood to apply to any of the above. Portions of this document that only apply to a subset of these devices will indicate so as appropriate.

The terms 3DoF and 6DoF are used throughout this document to describe the tracking capabilities of XR devices.

• A 3DoF device, short for "Three Degrees of Freedom", is one that can only track rotational movement. This is common in devices which rely exclusively on accelerometer and gyroscope readings to provide tracking. 3DoF devices do not respond to translational movements from the user, though they may employ algorithms to estimate translational changes based on modeling of the neck or arms.

• A 6DoF device, short for "Six Degrees of Freedom", is one that can track both rotation and translation, enabling precise 1:1 tracking in space. This typically requires some level of understanding of the user’s environment. That environmental understanding may be achieved via inside-out tracking, where sensors on the tracked device itself (such as cameras or depth sensors) are used to determine the device’s position, or outside-in tracking, where external devices placed in the user’s environment (like a camera or light emitting device) provides a stable point of reference against which the XR device can determine its position.

1.2. Application flow

2. Model

2.1. XR device

An XR device is a physical unit of hardware that can present immersive content to the user. Content is considered to be "immersive" if it produces visual, audio, haptic, or other sensory output that simulates or augments various aspects of the user’s environment. Most frequently this involves tracking the user’s motion in space and producing outputs that are synchronized to the user’s movement. On desktop clients, this is usually a headset peripheral. On mobile clients, it may represent the mobile device itself in conjunction with a viewer harness. It may also represent devices without stereo-presentation capabilities but with more advanced tracking.

An XR device has a list of supported modes (a list of strings) that contains the enumeration values of XRSessionMode that the XR device supports.

Each XR device has a set of granted features for each XRSessionMode in its list of supported modes, which is a set of feature descriptors which MUST be initially an empty set.

The user agent has a list of immersive XR devices (a list of XR device), which MUST be initially an empty list.

The user agent has an immersive XR device (null or XR device) which is initially null and represents the active XR device from the list of immersive XR devices. This object MAY live on a separate thread and be updated asynchronously.

The user agent MUST have a default inline XR device, which is an XR device that MUST contain "inline" in its list of supported modes. The default inline XR device MUST NOT report any pose information, and MUST NOT report XR input sources or events other than those created by pointer events.

Note: The default inline XR device exists purely as a convenience for developers, allowing them to use the same rendering and input logic for both inline and immersive content. The default inline XR device does not expose any information not already available to the developer through other mechanisms on the page (such as pointer events for input), it only surfaces those values in an XR-centric format.

The user agent MUST have a inline XR device, which is an XR device that MUST contain "inline" in its list of supported modes. The inline XR device MAY be the immersive XR device if the tracking it provides makes sense to expose to inline content or the default inline XR device otherwise.

Note: On phones, the inline XR device may report pose information derived from the phone’s internal sensors, such as the gyroscope and accelerometer. On desktops and laptops without similar sensors, the inline XR device will not be able to report a pose, and as such should fall back to the default inline XR device. In case the user agent is already running on an XR device, the inline XR device will be the same device, and may support multiple views. User consent must be given before any tracking or input features beyond what the default inline XR device exposes are provided.

The current values of list of immersive XR devices, inline XR device, and immersive XR device MAY live on a separate thread and be updated asynchronously. These objects SHOULD NOT be directly accessed in steps that are not running in parallel.

3. Initialization

3.1. navigator.xr

partial interface Navigator {
  [SecureContext, SameObject] readonly attribute XRSystem xr;
};

The xr attribute’s getter MUST return the XRSystem object that is associated with it.

3.2. XRSystem

[SecureContext, Exposed=Window] interface XRSystem : EventTarget {
  // Methods
  Promise<boolean> isSessionSupported(XRSessionMode mode);
  [NewObject] Promise<XRSession> requestSession(XRSessionMode mode, optional XRSessionInit options = {});

  // Events
  attribute EventHandler ondevicechange;
};

The user agent MUST create an XRSystem object when a Navigator object is created and associate it with that object.

An XRSystem object is the entry point to the API, used to query for XR features available to the user agent and initiate communication with XR hardware via the creation of XRSessions.

The user agent MUST be able to enumerate immersive XR devices attached to the system, at which time each available device is placed in the list of immersive XR devices. Subsequent algorithms requesting enumeration MUST reuse the cached list of immersive XR devices. Enumerating the devices should not initialize device tracking. After the first enumeration the user agent MUST begin monitoring device connection and disconnection, adding connected devices to the list of immersive XR devices and removing disconnected devices.

Note: The user agent is allowed to use any criteria it wishes to select an immersive XR device when the list of immersive XR devices contains multiple devices. For example, the user agent may always select the first item in the list, or provide settings UI that allows users to manage device priority. Ideally the algorithm used to select the default device is stable and will result in the same device being selected across multiple browsing sessions.

The ondevicechange attribute is an Event handler IDL attribute for the devicechange event type.

Note: The purpose of isSessionSupported() is not to report with perfect accuracy the user agent’s ability to create an XRSession, but to inform the page whether or not advertising the ability to create sessions of the given mode is advised. A certain level of false-positives are expected, even when user agent checks for the presence of the necessary hardware/software prior to resolving the method. (For example, even if the appropriate hardware is present it may have given exclusive access to another application at the time a session is requested.)

It is expected that most pages with XR content will call isSessionSupported() early in the document lifecycle. As such, calling isSessionSupported() SHOULD avoid displaying any modal or otherwise intrusive UI. Calling isSessionSupported() MUST NOT trigger device-selection UI, MUST NOT interfere with any running XR applications on the system, and MUST NOT cause XR-related applications to launch such as system trays or storefronts.

const supported = await navigator.xr.isSessionSupported('immersive-vr');
if (supported) {
  // 'immersive-vr' sessions may be supported.
  // Page should advertise support to the user.
} else {
  // 'immersive-vr' sessions are not supported.
}

The XRSystem object has a pending immersive session boolean, which MUST be initially false, an active immersive session, which MUST be initially null, and a list of inline sessions, which MUST be initially empty.

const xrSession = await navigator.xr.requestSession("immersive-vr");

3.3. XRSessionMode

The XRSessionMode enum defines the modes that an XRSession can operate in.

enum XRSessionMode {
  "inline",
  "immersive-vr",
  "immersive-ar"
};

• A session mode of inline indicates that the session’s output will be shown as an element in the HTML document. inline session content MUST be displayed in mono (i.e., with a single view). It MAY allow for viewer tracking. User agents MUST allow inline sessions to be created.

• A session mode of immersive-vr indicates that the session’s output will be given exclusive access to the immersive XR device display and that content is not intended to be integrated with the user’s environment.

• The behavior of the immersive-ar session mode is defined in the WebXR AR Module and MUST NOT be added to the immersive XR device’s list of supported modes unless the UA implements that module.

In this document, the term inline session is synonymous with an inline session and the term immersive session refers to either an immersive-vr or immersive-ar session.

Immersive sessions MUST provide some level of viewer tracking, and content MUST be shown at the proper scale relative to the user and/or the surrounding environment. Additionally, Immersive sessions MUST be given exclusive access to the immersive XR device, meaning that while the immersive session is "visible" the HTML document is not shown on the immersive XR device’s display, nor does content from any other source have exclusive access. Exclusive access does not prevent the user agent from overlaying its own UI, however this UI SHOULD be minimal.

Note: UA may choose to overlay content for accessibility or safety such as guardian boundaries, obstructions or the user’s hands when there are no alternative input sources.

Note: Future specifications or modules may expand the definition of immersive session to include additional session modes.

Note: Examples of ways exclusive access may be presented include stereo content displayed on a virtual reality headset.

Note: As an example of overlaid UI, the user agent or operating system in an immersive session may show notifications over the rendered content.

Note: While the HTML document is not shown on the immersive XR device’s display during an immersive session, it may still be shown on a separate display, e.g. when the user is entering the immersive session from a 2d browser on their computer tethered to their immersive XR device.

3.4. Feature Dependencies

Some features of an XRSession may not be universally available for a number of reasons, among which is the fact not all XR devices can support the full set of features. Another consideration is that some features expose sensitive information which may require a clear signal of user intent before functioning.

Since it is a poor user experience to initialize the underlying XR platform and create an XRSession only to immediately notify the user that the applications cannot function correctly, developers can indicate required features by passing an XRSessionInit dictionary to requestSession(). This will block the creation of the XRSession if any of the required features are unavailable due to device limitations or in the absence of a clear signal of user intent to expose sensitive information related to the feature.

Additionally, developers are encouraged to design experiences which progressively enhance their functionality when run on more capable devices. Optional features which the experience does not require but will take advantage of when available must also be indicated in an XRSessionInit dictionary to ensure that user intent can be determined before enabling the feature if necessary.

dictionary XRSessionInit {
  sequence<DOMString> requiredFeatures;
  sequence<DOMString> optionalFeatures;
};

The requiredFeatures array contains any Required features for the experience. If any value in the list is not a recognized feature descriptor the XRSession will not be created. If any feature listed in the requiredFeatures array is not supported by the XR device or, if necessary, has not received a clear signal of user intent the XRSession will not be created.

The optionalFeatures array contains any Optional features for the experience. If any value in the list is not a recognized feature descriptor it will be ignored. Features listed in the optionalFeatures array will be enabled if supported by the XR device and, if necessary, given a clear signal of user intent, but will not block creation of the XRSession if absent.

Values given in the feature lists are considered a valid feature descriptor if the value is one of the following:

• The string representation of any XRReferenceSpaceType enum value

Future iterations of this specification and additional modules may expand the list of accepted feature descriptors.

Note: If a feature needs additional initialization, XRSessionInit should be extended with a new field for that feature.

Depending on the XRSessionMode requested, certain feature descriptors are added to the requiredFeatures or optionalFeatures lists by default. The following table describes the default features associated with each session type and feature list:

The combined list of feature descriptors given by the requiredFeatures and optionalFeatures are collectively considered the requested features for an XRSession.

Some feature descriptors, when present in the requested features list, are subject to permissions policy and/or requirements that user intent to use the feature is well understood, via either explicit consent or implicit consent. The following table describes the feature requirements that must be satisfied prior to being enabled:

Note: "local" is always included in the requested features of immersive sessions as a default feature, and as such immersive sessions always need to obtain explicit consent or implicit consent.

Requested features can only be enabled for a session if the XR device is capable of supporting the feature, which means that the feature is known to be supported by the XR device in some configurations, even if the current configuration has not yet been verified as supporting the feature. The user agent MAY apply more rigorous constraints if desired in order to yield a more consistent user experience.

Note: For example, several VR devices support either configuring a safe boundary for the user to move around within or skipping boundary configuration and operating in a mode where the user is expected to stand in place. Such a device can be considered to be capable of supporting "bounded-floor" XRReferenceSpaces even if they are currently not configured with safety boundaries, because it’s expected that the user could configure the device appropriately if the experience required it. This is to allow user agents to avoid fully initializing the XR device or waiting for the user’s environment to be recognized prior to resolving the requested features if desired. If, however, the user agent knows the boundary state at the time the session is requested without additional initialization it may choose to reject the "bounded-floor" feature if the safety boundary is not already configured.

4. Session

4.1. XRSession

Any interaction with XR hardware is done via an XRSession object, which can only be retrieved by calling requestSession() on the XRSystem object. Once a session has been successfully acquired, it can be used to poll the viewer pose, query information about the user’s environment, and present imagery to the user.

The user agent, when possible, SHOULD NOT initialize device tracking or rendering capabilities until an XRSession has been acquired. This is to prevent unwanted side effects of engaging the XR systems when they’re not actively being used, such as increased battery usage or related utility applications from appearing when first navigating to a page that only wants to test for the presence of XR hardware in order to advertise XR features. Not all XR platforms offer ways to detect the hardware’s presence without initializing tracking, however, so this is only a strong recommendation.

enum XRVisibilityState {
  "visible",
  "visible-blurred",
  "hidden",
};

[SecureContext, Exposed=Window] interface XRSession : EventTarget {
  // Attributes
  readonly attribute XRVisibilityState visibilityState;
  readonly attribute float? frameRate;
  readonly attribute Float32Array? supportedFrameRates;
  [SameObject] readonly attribute XRRenderState renderState;
  [SameObject] readonly attribute XRInputSourceArray inputSources;
  [SameObject] readonly attribute XRInputSourceArray trackedSources;
  readonly attribute FrozenArray<DOMString> enabledFeatures;
  readonly attribute boolean isSystemKeyboardSupported;

  // Methods
  undefined updateRenderState(optional XRRenderStateInit state = {});
  Promise<undefined> updateTargetFrameRate(float rate);
  [NewObject] Promise<XRReferenceSpace> requestReferenceSpace(XRReferenceSpaceType type);

  unsigned long requestAnimationFrame(XRFrameRequestCallback callback);
  undefined cancelAnimationFrame(unsigned long handle);

  Promise<undefined> end();

  // Events
  attribute EventHandler onend;
  attribute EventHandler oninputsourceschange;
  attribute EventHandler onselect;
  attribute EventHandler onselectstart;
  attribute EventHandler onselectend;
  attribute EventHandler onsqueeze;
  attribute EventHandler onsqueezestart;
  attribute EventHandler onsqueezeend;
  attribute EventHandler onvisibilitychange;
  attribute EventHandler onframeratechange;
};

Each XRSession has a mode, which is one of the values of XRSessionMode.

Each XRSession has an animation frame, which is an XRFrame initialized with active set to false, animationFrame set to true, and session set to the XRSession.

Each XRSession has a set of granted features, which is a set of DOMStrings corresponding to the feature descriptors that have been granted to the XRSession.

The enabledFeatures attribute returns the features in the set of granted features as a new array of DOMStrings.

The isSystemKeyboardSupported attribute indicates that the XRSystem has the ability to display the system keyboard while the XRSession is active. If isSystemKeyboardSupported is true, Web APIs that would trigger the overlay keyboard (such as focus) will show the system keyboard. The XRSession MUST set the visibility state of the XRSession to "visible-blurred" while the keyboard is shown.

A number of different circumstances may shut down the session, which is permanent and irreversible. Once a session has been shut down the only way to access the XR device’s tracking or rendering capabilities again is to request a new session. Each XRSession has an ended boolean, initially set to false, that indicates if it has been shut down.

Each XRSession has an active render state which is a new XRRenderState, and a pending render state, which is an XRRenderState which is initially null.

The renderState attribute returns the XRSession’s active render state.

Each XRSession has a minimum inline field of view and a maximum inline field of view, defined in radians. The values MUST be determined by the user agent and MUST fall in the range of 0 to PI.

Each XRSession has a minimum near clip plane and a maximum far clip plane, defined in meters. The values MUST be determined by the user agent and MUST be non-negative. The minimum near clip plane SHOULD be less than 0.1. The maximum far clip plane SHOULD be greater than 1000.0 (and MAY be infinite).

If the XR Compositor changes the nominal frame rate for any reason (for example during a "visible-blurred" event), it SHOULD use the internal target framerate once the event that caused the frame rate change has ended.

Each XRSession has a list of active XR input sources (a list of XRInputSource) and a list of active XR tracked sources (a list of XRInputSource) which MUST both be initially an empty list.

Each XRSession has an XR device, which is an XR device set at initialization.

The inputSources attribute returns the XRSession’s list of active XR input sources.

The trackedSources attribute returns the XRSession’s list of active XR tracked sources.

The user agent MUST monitor any XR input sources associated with the XR device, including detecting when XR input sources are added, removed, or changed.

Each XRSession has a promise resolved flag, initially false.

NOTE: The purpose of this flag is to ensure that the add input source, remove input source, and change input source algorithms do not run until the user code actually has had a chance to attach event listeners. Implementations may not need this flag if they simply choose to start listening for input source changes after the session resolves.

Each XRSession has a visibility state value, which is an enum. For inline sessions the visibility state MUST mirror the Document’s visibilityState. For immersive sessions the visibility state MUST be set to whichever of the following values best matches the state of session.

• A state of visible indicates that imagery rendered by the XRSession can be seen by the user and requestAnimationFrame() callbacks are processed at the XR device’s native refresh rate. Input is processed by the XRSession normally.

• A state of visible-blurred indicates that imagery rendered by the XRSession may be seen by the user, but is not the primary focus. requestAnimationFrame() callbacks MAY be throttled. Input is not processed by the XRSession.

• A state of hidden indicates that imagery rendered by the XRSession cannot be seen by the user. requestAnimationFrame() callbacks will not be processed until the visibility state changes. Input is not processed by the XRSession.

The visibilityState attribute returns the XRSession’s visibility state. The onvisibilitychange attribute is an Event handler IDL attribute for the visibilitychange event type.

The visibility state MAY be changed by the user agent at any time other than during the processing of an XR animation frame, and the user agent SHOULD monitor the XR platform when possible to observe when session visibility has been affected external to the user agent and update the visibility state accordingly.

Note: The XRSession’s visibility state does not necessarily imply the visibility of the HTML document. Depending on the system configuration the page may continue to be visible while an immersive session is active. (For example, a headset connected to a PC may continue to display the page on the monitor while the headset is viewing content from an immersive session.) Developers should continue to rely on the Page Visibility to determine page visibility.

Note: The XRSession’s visibility state does not affect or restrict mouse behavior on tethered sessions where 2D content is still visible while an immersive session is active. Content should consider using the [pointerlock] API if it wishes to have stronger control over mouse behavior.

In an XRSystem, there are several definitions which can describe a frame rate:

• The nominal frame rate: the rate at which the XRSystem is asking the experience to render frames to maintain nominal performance. Experiences that miss frames may not end up actually getting calls to requestAnimationFrame() this many times per second, but that is what the XRSystem is aiming to achieve.

• The effective frame rate: a performance measurement of how many calls to requestAnimationFrame() the experience is actually managing to process each second. This will fluctuate based on the experience hitting or missing the XRSystem’s frame timing.

• The target frame rate: the experience’s hint to the XRSystem on what nominal frame rate it prefers to target.

• The display frame rate: the actual rate at which frames are drawn to the physical display, which MAY be derived from the experience’s nominal frame rate. This is a hardware implementation detail that is not exposed to the experience.

Each XRSession MAY have an internal target frameRate which is the target frame rate.

Each XRSession MAY have an internal nominal frameRate which is the nominal frame rate. If the effective frame rate is lower than the nominal frame rate, the XR Compositor MAY use reprojection or other techniques to improve the experience. It is optional and MUST NOT be present for inline sessions.

The frameRate attribute reflects the internal nominal framerate. If the XRSession has no internal nominal framerate, return null.

The onframeratechange attribute is an Event handler IDL attribute for the frameratechange event type. If XRSession’s nominal frame rate is changed for any reason, it MUST apply the nominal frame rate with the new nominal frame rate and the XRSession.

The supportedFrameRates attribute returns a list of supported target frame rate values. This attribute is optional and MUST NOT be present for inline sessions or for an XRSystem that doesn’t let the author control the frame rate. If the XRSession supports the supportedFrameRates attribute, it also MUST support frameRate.

Each XRSession has a viewer reference space, which is an XRReferenceSpace of type "viewer" with an identity transform origin offset.

Each XRSession has a list of views, which is a list of views corresponding to the views provided by the XR device. If the XRSession’s renderState’s composition enabled boolean is set to false the list of views MUST contain a single view. The list of views is immutable during the XRSession and MUST contain any views that may be surfaced during the session, including secondary views that may not initially be active.

The onend attribute is an Event handler IDL attribute for the end event type.

The oninputsourceschange attribute is an Event handler IDL attribute for the inputsourceschange event type.

The onselectstart attribute is an Event handler IDL attribute for the selectstart event type.

The onselectend attribute is an Event handler IDL attribute for the selectend event type.

The onselect attribute is an Event handler IDL attribute for the select event type.

The onsqueezestart attribute is an Event handler IDL attribute for the squeezestart event type.

The onsqueezeend attribute is an Event handler IDL attribute for the squeezeend event type.

The onsqueeze attribute is an Event handler IDL attribute for the squeeze event type.

4.2. XRRenderState

An XRRenderState represents a set of configurable values which affect how an XRSession’s output is composited. The active render state for a given XRSession can only change between frame boundaries, and updates can be queued up via updateRenderState().

dictionary XRRenderStateInit {
  double depthNear;
  double depthFar;
  boolean passthroughFullyObscured;
  double inlineVerticalFieldOfView;
  XRWebGLLayer? baseLayer;
  sequence<XRLayer>? layers;
};

[SecureContext, Exposed=Window] interface XRRenderState {
  readonly attribute double depthNear;
  readonly attribute double depthFar;
  readonly attribute boolean? passthroughFullyObscured;
  readonly attribute double? inlineVerticalFieldOfView;
  readonly attribute XRWebGLLayer? baseLayer;
};

Each XRRenderState has a output canvas, which is an HTMLCanvasElement initially set to null. The output canvas is the DOM element that will display any content rendered for an "inline" XRSession.

Each XRRenderState also has a composition enabled boolean, which is initially true. The XRRenderState is considered to have composition enabled if rendering commands are performed against a surface provided by the API and displayed by the XR Compositor. If rendering is performed for an "inline" XRSession in such a way that it is directly displayed into an output canvas, the XRRenderState’s composition enabled flag MUST be false.

Note: At this point the XRRenderState will only have an output canvas if it has composition enabled set to false, but future versions of the specification are likely to introduce methods for setting output canvases that support more advanced uses like mirroring and layer compositing that will require composition.

The depthNear attribute defines the distance, in meters, of the near clip plane from the viewer. The depthFar attribute defines the distance, in meters, of the far clip plane from the viewer.

depthNear and depthFar are used in the computation of the projectionMatrix of XRViews. When the projectionMatrix is used during rendering, only geometry with a distance to the viewer that falls between depthNear and depthFar will be drawn. They also determine how the values of an XRWebGLLayer depth buffer are interpreted. depthNear MAY be greater than depthFar.

Note: Typically when constructing a perspective projection matrix for rendering the developer specifies the viewing frustum and the near and far clip planes. When displaying to an immersive XR device the correct viewing frustum is determined by some combination of the optics, displays, and cameras being used. The near and far clip planes, however, may be modified by the application since the appropriate values depend on the type of content being rendered.

The passthroughFullyObscured attribute is a hint to the XRSystem from the author to indicate that they intend to completely cover the viewport with virtual content. The author SHOULD set this flag back to false once the viewport is no longer covered by opaque pixels.

NOTE: the XRSystem MAY use this as a hint to temporarily disable passthrough. On devices with see-through optics, the user will continue to see their environment and this flag will have no effect.

The inlineVerticalFieldOfView attribute defines the default vertical field of view in radians used when computing projection matrices for "inline" XRSessions. The projection matrix calculation also takes into account the aspect ratio of the output canvas. This value MUST be null for immersive sessions.

The baseLayer attribute defines an XRWebGLLayer which the XR compositor will obtain images from.

4.3. Animation Frames

The primary way an XRSession provides information about the tracking state of the XR device is via callbacks scheduled by calling requestAnimationFrame() on the XRSession instance.

callback XRFrameRequestCallback = undefined (DOMHighResTimeStamp time, XRFrame frame);

Each XRFrameRequestCallback object has a cancelled boolean initially set to false.

Each XRSession has a list of animation frame callbacks, which is initially empty, a list of currently running animation frame callbacks, which is also initially empty, and an animation frame callback identifier, which is a number which is initially zero.

The behavior of the Window interface’s requestAnimationFrame() method is not changed by the presence of any active XRSession, nor does calling requestAnimationFrame() on any XRSession interact with Window’s requestAnimationFrame() in any way. An active immersive session MAY affect the rendering opportunity of a browsing context if it causes the page to be obscured. If the 2D browser view is visible during an active immersive session (i.e., when the sesson is running on a tethered headset), the timing of callbacks run with Window’s requestAnimationFrame() and requestIdleCallback() MAY NOT coincide with that of the session’s requestAnimationFrame() and should not be relied upon by the user for rendering XR content.

Note: User agents may wish to display a warning to the developer console if XRSession’s requestAnimationFrame() is called during callbacks scheduled via Window’s requestAnimationFrame(), as these callbacks are not guaranteed to occur if the active immersive session affects the rendering opportunity of the browsing context, and may not have the correct timing even if they run.

let xrSession = null;

function onWindowAnimationFrame(time) {
  window.requestAnimationFrame(onWindowAnimationFrame);

  // This may be called while an immersive session is running on some devices,
  // such as a desktop with a tethered headset. To prevent two loops from
  // rendering in parallel, skip drawing in this one until the session ends.
  if (!xrSession) {
    renderFrame(time, null);
  }
}

// The window animation loop can be started immediately upon the page loading.
window.requestAnimationFrame(onWindowAnimationFrame);

function onXRAnimationFrame(time, xrFrame) {
  xrSession.requestAnimationFrame(onXRAnimationFrame);
  renderFrame(xrFrame.predictedDisplayTime, xrFrame);
}

function renderFrame(time, xrFrame) {
  // Shared rendering logic.
}

// Assumed to be called by a user gesture event elsewhere in code.
async function startXRSession() {
  xrSession = await navigator.xr.requestSession('immersive-vr');
  xrSession.addEventListener('end', onXRSessionEnded);
  // Do necessary session setup here.
  // Begin the session’s animation loop.
  xrSession.requestAnimationFrame(onXRAnimationFrame);
}

function onXRSessionEnded() {
  xrSession = null;
}

4.4. The XR Compositor

The user agent MUST maintain an XR Compositor which handles presentation to the XR device and frame timing. The compositor MUST use an independent rendering context whose state is isolated from that of any graphics contexts created by the document. The compositor MUST prevent the page from corrupting the compositor state or reading back content from other pages or applications. The compositor MUST also run in separate thread or processes to decouple performance of the page from the ability to present new imagery to the user at the appropriate framerate. The compositor MAY composite additional device or user agent UI over rendered content, like device menus.

Note: Future extensions to this spec may utilize the compositor to composite multiple layers coming from the same page as well.

5. Frame Loop

5.1. XRFrame

An XRFrame represents a snapshot of the state of all of the tracked objects for an XRSession. Applications can acquire an XRFrame by calling requestAnimationFrame() on an XRSession with an XRFrameRequestCallback. When the callback is called it will be passed an XRFrame. Events which need to communicate tracking state, such as the select event, will also provide an XRFrame.

[SecureContext, Exposed=Window] interface XRFrame {
  [SameObject] readonly attribute XRSession session;
  readonly attribute DOMHighResTimeStamp predictedDisplayTime;

  XRViewerPose? getViewerPose(XRReferenceSpace referenceSpace);
  XRPose? getPose(XRSpace space, XRSpace baseSpace);
};

Each XRFrame has an active boolean which is initially set to false, and an animationFrame boolean which is initially set to false.

The session attribute returns the XRSession that produced the XRFrame.

For an immersive session the predictedDisplayTime attribute MUST return the DOMHighResTimeStamp corresponding to the average point in time this XRFrame is expected to be displayed on the devices' display. For an "inline" XRSession, predictedDisplayTime MUST return the same value as the timestamp passed to the XRFrameRequestCallback.

Each XRFrame represents the state of all tracked objects for a given time, and either stores or is able to query concrete information about this state at the time.

A frame update is an algorithm that can be run given an XRFrame, which is intended to be run each XRFrame.

Every XRSession has a list of frame updates, which is a list of frame updates, initially the empty list.

NOTE: This spec does not define any frame updates, but other specifications may add some.

6. Spaces

A core feature of the WebXR Device API is the ability to provide spatial tracking. Spaces are the interface that enable applications to reason about how tracked entities are spatially related to the user’s physical environment and each other.

6.1. XRSpace

An XRSpace represents a virtual coordinate system with an origin that corresponds to a physical location. Spatial data that is requested from the API or given to the API is always expressed in relation to a specific XRSpace at the time of a specific XRFrame. Numeric values such as pose positions are coordinates in that space relative to its origin. The interface is intentionally opaque.

[SecureContext, Exposed=Window] interface XRSpace : EventTarget {

};

Each XRSpace has a session which is set to the XRSession that created the XRSpace.

Each XRSpace has a native origin which is a position and orientation in space. The XRSpace’s native origin may be updated by the XR device’s underlying tracking system, and different XRSpaces may define different semantics as to how their native origins are tracked and updated.

Each XRSpace has an effective origin, which is the basis of the XRSpace’s coordinate system.

The transform from the effective space to the native origin’s space is defined by an origin offset, which is an XRRigidTransform initially set to an identity transform. In other words, the effective origin can be obtained by multiplying origin offset and the native origin.

The effective origin of an XRSpace can only be observed in the coordinate system of another XRSpace as an XRPose, returned by an XRFrame’s getPose() method. The spatial relationship between XRSpaces MAY change between XRFrames.

6.2. XRReferenceSpace

An XRReferenceSpace is one of several common XRSpaces that applications can use to establish a spatial relationship with the user’s physical environment.

XRReferenceSpaces are generally expected to remain static for the duration of the XRSession, with the most common exception being mid-session reconfiguration by the user. The native origin for every XRReferenceSpace describes a coordinate system where +X is considered "Right", +Y is considered "Up", and -Z is considered "Forward".

enum XRReferenceSpaceType {
  "viewer",
  "local",
  "local-floor",
  "bounded-floor",
  "unbounded"
};

[SecureContext, Exposed=Window]
interface XRReferenceSpace : XRSpace {
  [NewObject] XRReferenceSpace getOffsetReferenceSpace(XRRigidTransform originOffset);

  attribute EventHandler onreset;
};

Each XRReferenceSpace has a type, which is an XRReferenceSpaceType.

An XRReferenceSpace is most frequently obtained by calling requestReferenceSpace(), which creates an instance of an XRReferenceSpace (or an interface extending it) if the XRReferenceSpaceType enum value passed into the call is supported. The type indicates the tracking behavior that the reference space will exhibit:

• Passing a type of viewer creates an XRReferenceSpace instance. It represents a tracking space with a native origin which tracks the position and orientation of the viewer. Every XRSession MUST support "viewer" XRReferenceSpaces.

• Passing a type of local creates an XRReferenceSpace instance. It represents a tracking space with a native origin near the viewer at the time of creation. The exact position and orientation will be initialized based on the conventions of the underlying platform. When using this reference space the user is not expected to move beyond their initial position much, if at all, and tracking is optimized for that purpose. For devices with 6DoF tracking, local reference spaces should emphasize keeping the origin stable relative to the user’s environment.

• Passing a type of local-floor creates an XRReferenceSpace instance. It represents a tracking space with a native origin at the floor in a safe position for the user to stand. The Y axis equals 0 at floor level, with the X and Z position and orientation initialized based on the conventions of the underlying platform. If the floor level isn’t known it MUST be estimated, with some estimated floor level. If the estimated floor level is determined with a non-default value, it MUST be rounded sufficiently to prevent fingerprinting. When using this reference space the user is not expected to move beyond their initial position much, if at all, and tracking is optimized for that purpose. For devices with 6DoF tracking, local-floor reference spaces should emphasize keeping the origin stable relative to the user’s environment.

  Note: If the floor level of a "local-floor" reference space is adjusted to prevent fingerprinting, rounded to the nearest 1cm is suggested.

• Passing a type of bounded-floor creates an XRBoundedReferenceSpace instance. It represents a tracking space with its native origin at the floor, where the user is expected to move within a pre-established boundary, given as the boundsGeometry. Tracking in a bounded-floor reference space is optimized for keeping the native origin and boundsGeometry stable relative to the user’s environment.

• Passing a type of unbounded creates an XRReferenceSpace instance. It represents a tracking space where the user is expected to move freely around their environment, potentially even long distances from their starting point. Tracking in an unbounded reference space is optimized for stability around the user’s current position, and as such the native origin may drift over time.

Note: It is assumed that the conventions of the underlying platform regarding Y axes of the reference spaces stay consistent across different types of XRReferenceSpaces. In other words, if an XR system supports multiple reference spaces, their Y axes will be parallel to each other and point in the same direction for the duration of the XRSession in which they were created. This does not apply to "viewer", which does not rely on the conventions of the underlying platform for its orientation. "unbounded" reference spaces should align their Y axes with other reference spaces when their origins are nearby, but may deviate if the user moves over large distances.

Devices that support "local" reference spaces MUST support "local-floor" reference spaces, through emulation if necessary, and vice versa.

The onreset attribute is an Event handler IDL attribute for the reset event type.

Note: It’s expected that some applications will use getOffsetReferenceSpace() to implement scene navigation controls based on mouse, keyboard, touch, or gamepad input. This will result in getOffsetReferenceSpace() being called frequently, at least once per-frame during periods of active input. As a result UAs are strongly encouraged to make the creation of new XRReferenceSpaces with getOffsetReferenceSpace() a lightweight operation.

6.3. XRBoundedReferenceSpace

XRBoundedReferenceSpace extends XRReferenceSpace to include boundsGeometry, indicating the pre-configured boundaries of the user’s space.

[SecureContext, Exposed=Window]
interface XRBoundedReferenceSpace : XRReferenceSpace {
  readonly attribute FrozenArray<DOMPointReadOnly> boundsGeometry;
};

The origin of an XRBoundedReferenceSpace MUST be positioned at the floor, such that the Y axis equals 0 at floor level. The X and Z position and orientation are initialized based on the conventions of the underlying platform, typically expected to be near the center of the room facing in a logical forward direction.

Note: Other XR platforms sometimes refer to the type of tracking offered by a bounded-floor reference space as "room scale" tracking. An XRBoundedReferenceSpace is not intended to describe multi-room spaces, areas with uneven floor levels, or very large open areas. Content that needs to handle those scenarios should use an unbounded reference space.

Each XRBoundedReferenceSpace has a native bounds geometry describing the border around the XRBoundedReferenceSpace, which the user can expect to safely move within. The polygonal boundary is given as an array of DOMPointReadOnlys, which represents a loop of points at the edges of the safe space. The points describe offsets from the native origin in meters. Points MUST be given in a clockwise order as viewed from above, looking towards the negative end of the Y axis. The y value of each point MUST be 0 and the w value of each point MUST be 1. The bounds can be considered to originate at the floor and extend infinitely high. The shape it describes MAY be convex or concave.

Each point in the native bounds geometry MUST be limited to a reasonable distance from the reference space’s native origin.

Note: It is suggested that points of the native bounds geometry be limited to 15 meters from the native origin in all directions.

Each point in the native bounds geometry MUST also be quantized sufficiently to prevent fingerprinting. For user’s safety, quantized points values MUST NOT fall outside the bounds reported by the platform.

Note: It is suggested that points of the native bounds geometry be quantized to the nearest 5cm.

The boundsGeometry attribute is an array of DOMPointReadOnlys such that each entry is equal to the entry in the XRBoundedReferenceSpace’s native bounds geometry premultiplied by the inverse of the origin offset. In other words, it provides the same border in XRBoundedReferenceSpace coordinates relative to the effective origin.

If the native bounds geometry is temporarily unavailable, which may occur for several reasons such as during XR device initialization, extended periods of tracking loss, or movement between pre-configured spaces, the boundsGeometry MUST report an empty array.

Note: Bounded reference spaces may be returned if the boundaries or floor height have not been resolved at the time of the reference space request, but the XR device is known to support them.

Note: Content should not require the user to move beyond the boundsGeometry. It is possible for the user to move beyond the bounds if their physical surroundings allow for it, resulting in position values outside of the polygon they describe. This is not an error condition and should be handled gracefully by page content.

Note: Content generally should not provide a visualization of the boundsGeometry, as it’s the user agent’s responsibility to ensure that safety critical information is provided to the user.

7. Views

7.1. XRViewGeometry

Objects including the XRViewGeometry interface mixin represent either a display used by an XR device to present imagery to the user or a sensor used to collect visual information about the real world. These objects contain a view geometry.

A view geometry corresponds to the set of intrinsics and extrinsics used to translate between a point in the viewer reference space’s coordinate system, and in the containing object’s screen space.

A view geometry has a containing object which is the physical piece of hardware that the view geometry contains data for.

A view geometry’s containing object has an associated screen space, which is described as the 2D plane that this containing object either reads data from or renders data to.

A view geometry has an associated view offset, which is an XRRigidTransform describing the position and orientation of the containing object in the viewer reference space’s coordinate system.

NOTE: There are no constraints on what the view offset might be, and views are allowed to have differing orientations. This can crop up in head-mounted devices with eye displays centered at an angle, and it can also surface itself in more extreme cases like CAVE rendering. Techniques like z-sorting and culling may need to be done per-eye because of this.

A view geometry has an associated projection matrix which is a matrix describing the projection to be used when rendering to the containing object provided by the underlying XR device. The projection matrix MAY include transformations such as shearing that prevent the projection from being accurately described by a simple frustum.

Note: The inverse of this matrix is suitable for "reading" pixels out of the screen space and translating them back to the coordinate system with the containing object as the origin.

[SecureContext, Exposed=Window] interface mixin XRViewGeometry {
  readonly attribute Float32Array projectionMatrix;
  [SameObject] readonly attribute XRRigidTransform transform;
};

Each XRViewGeometry has an associated internal projection matrix which stores the projection matrix of its containing object. It is initially null.

Note: The transform can be used to position camera objects in many rendering libraries. If a more traditional view matrix is needed by the application one can be retrieved by calling transform.inverse.matrix.

The projectionMatrix attribute is the projection matrix of the underlying view geometry. It is strongly recommended that applications use this matrix without modification or decomposition. Failure to use the provided projection matrices when rendering may cause the presented frame to be distorted or badly aligned, resulting in varying degrees of user discomfort. This attribute MUST be computed by obtaining the projection matrix for the XRViewGeometry.

The transform attribute is the XRRigidTransform of the object. It represents the position and orientation of the object in the XRReferenceSpace used to obtain the object.

7.2. XRView

An XRView describes a single view into an XR scene for a given frame.

A view corresponds to a display or portion of a display used by an XR device to present imagery to the user. They are used to retrieve all the information necessary to render content that is well aligned to the view’s physical output properties, including the field of view, eye offset, and other optical properties. Views may cover overlapping regions of the user’s vision. No guarantee is made about the number of views any XR device uses or their order, nor is the number of views required to be constant for the duration of an XRSession.

A view has an associated eye which is an XREye describing which eye this view is expected to be shown to. If the view does not have an intrinsically associated eye (the display is monoscopic, for example) this value MUST be set to "none".

A view has an active flag that may change through the lifecycle of an XRSession. Primary views MUST always have the active flag set to true.

Note: Many HMDs will request that content render two views, one for the left eye and one for the right, while most magic window devices will only request one view, but applications should never assume a specific view configuration. For example: A magic window device may request two views if it is capable of stereo output, but may revert to requesting a single view for performance reasons if the stereo output mode is turned off. Similarly, HMDs may request more than two views to facilitate a wide field of view or displays of different pixel density.

A view has an internal viewport modifiable flag that indicates if the viewport scale can be changed by a requestViewportScale() call at this point in the session. It is set to true at the start of an animation frame, and set to false when getViewport() is called.

A view has an internal requested viewport scale value that represents the requested viewport scale for this view. It is initially set to 1.0, and can be modified by the requestViewportScale() method if the system supports dynamic viewport scaling.

A view has an internal current viewport scale value that represents the current viewport scale for this view as used internally by the system. It is initially set to 1.0. It is updated to match the requested viewport scale when the viewport change is successfully applied by a getViewport() call.

A view has an reference space, which is the XRReferenceSpace space used to obtain this view in getViewerPose()

Note: Dynamic viewport scaling allows applications to render to a subset of the full-sized viewport using a scale factor that can be changed every animation frame. This is intended to be efficiently modifiable on a per-frame basis without reallocation. For correct rendering, it’s essential that the XR system and application agree on the active viewport. An application can call requestViewportScale() for an XRView multiple times within a single animation frame, but the requested scale does not take effect until the application calls getViewport() for that view. The first getViewport call in an animation frame applies the change (taking effect immediately for the current animation frame), locks in the view’s current scaled viewport for the remainder of this animation frame, and sets the scale as the new default for future animation frames. Optionally, the system can provide a suggested value through the recommendedViewportScale attribute based on internal performance heuristics and target framerates.

enum XREye {
  "none",
  "left",
  "right"
};

[SecureContext, Exposed=Window] interface XRView {
  readonly attribute XREye eye;
  readonly attribute unsigned long index;
  readonly attribute double? recommendedViewportScale;

  undefined requestViewportScale(double? scale);
};

XRView includes XRViewGeometry;

The transform is given in it’s reference space.

The eye attribute describes the eye of the underlying view. This attribute’s primary purpose is to ensure that pre-rendered stereo content can present the correct portion of the content to the correct eye.

The index attribute describes the offet of this XRView when it is return in the views array by getViewerPose().

The optional recommendedViewportScale attribute contains a UA-recommended viewport scale value that the application can use for a requestViewportScale() call to configure dynamic viewport scaling. It is null if the system does not implement a heuristic or method for determining a recommended scale. If not null, the value MUST be a numeric value greater than 0.0 and less than or equal to 1.0, and MUST be quantized to avoid providing detailed performance or GPU utilization data.

Note: It is suggested to quantize the recommended viewport scale by rounding it to the nearest value from a short list of possible scale values, and using hysteresis to avoid instant changes when close to a boundary value. (This also helps avoid rapidly oscillating scale values which can be visually distracting or uncomfortable.)

Each XRView has an associated session which is the XRSession that produced it.

Each XRView has an associated frame which is the XRFrame that produced it.

Each XRView has an associated underlying view which is the underlying view that it represents.

7.3. Primary and Secondary Views

A view is a primary view when rendering to it is necessary for an immersive experience. Primary views MUST be active for the entire duration of the XRSession.

A view is a secondary view when it is possible for content to choose to not render to it and still produce a working immersive experience. When content chooses to not render to these views, the user agent MAY be able to reconstruct them via reprojection. Secondary views MUST NOT be active unless the "secondary-views" feature is enabled.

To provide for this, user agents that expose secondary views MUST support an "secondary-views" feature descriptor as a hint. Content enabling this feature is expected to:

• Handle any nonzero number of views in the views array.

• Handle the existence of multiple views that have the same eye.

• Handle the size of the views array changing from frame to frame. This can happen when video capture is enabled, for example

When "secondary-views" is enabled, the user agent MAY surface any secondary views the device supports to the XRSession, when necessary. The user agent MUST NOT use reprojection to reconstruct secondary views in such a case, and instead rely on whatever the content decides to render.

Note: We recommend content use optionalFeatures to enable "secondary-views" to ensure maximum compatibility.

If secondary views have lower underlying frame rates, the XRSession MAY choose to do one or more of the following:

• Lower the overall frame rate of the application while the secondary views are active.

• Surface secondary views in the views array only for some of the frames. Implementations doing this SHOULD NOT have frames where the primary views are not present.

• Silently discard rendered content for secondary views during some of the frames.

7.4. XRViewport

An XRViewport object describes a viewport, or rectangular region, of a graphics surface.

[SecureContext, Exposed=Window] interface XRViewport {
  readonly attribute long x;
  readonly attribute long y;
  readonly attribute long width;
  readonly attribute long height;
};

The x and y attributes define an offset from the surface origin and the width and height attributes define the rectangular dimensions of the viewport.

The exact interpretation of the viewport values depends on the conventions of the graphics API the viewport is associated with:

• When used with an XRWebGLLayer the x and y attributes specify the lower left corner of the viewport rectangle, in pixels, with the viewport rectangle extending width pixels to the right of x and height pixels above y. The values can be passed to the WebGL viewport function directly.

xrSession.requestAnimationFrame((time, xrFrame) => {
  const viewer = xrFrame.getViewerPose(xrReferenceSpace);

  gl.bindFramebuffer(xrWebGLLayer.framebuffer);
  for (xrView of viewer.views) {
    let xrViewport = xrWebGLLayer.getViewport(xrView);
    gl.viewport(xrViewport.x, xrViewport.y, xrViewport.width, xrViewport.height);

    // WebGL draw calls will now be rendered into the appropriate viewport.
  }
});

8. Geometric Primitives

8.1. Matrices

WebXR provides various transforms in the form of matrices. WebXR uses the WebGL conventions when communicating matrices, in which 4x4 matrices are given as 16 element Float32Arrays with column major storage, and are applied to column vectors by premultiplying the matrix from the left. They may be passed directly to WebGL’s uniformMatrix4fv function, used to create an equivalent DOMMatrix, or used with a variety of third party math libraries.

[a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11, a12, a13, a14, a15]

{x:X, y:Y, z:Z, w:1}

a0 a4 a8  a12  *  X  =  a0 * X + a4 * Y +  a8 * Z + a12
a1 a5 a9  a13     Y     a1 * X + a5 * Y +  a9 * Z + a13
a2 a6 a10 a14     Z     a2 * X + a6 * Y + a10 * Z + a14
a3 a7 a11 a15     1     a3 * X + a7 * Y + a11 * Z + a15

8.2. Normalization

There are several algorithms which call for a vector or quaternion to be normalized, which means to scale the components to have a collective magnitude of 1.0.

8.3. XRRigidTransform

An XRRigidTransform is a transform described by a position and orientation. When interpreting an XRRigidTransform the orientation is always applied prior to the position.

An XRRigidTransform contains an internal matrix which is a matrix.

[SecureContext, Exposed=Window]
interface XRRigidTransform {
  constructor(optional DOMPointInit position = {}, optional DOMPointInit orientation = {});
  [SameObject] readonly attribute DOMPointReadOnly position;
  [SameObject] readonly attribute DOMPointReadOnly orientation;
  readonly attribute Float32Array matrix;
  [SameObject] readonly attribute XRRigidTransform inverse;
};

The position attribute is a 3-dimensional point, given in meters, describing the translation component of the transform. The position’s w attribute MUST be 1.0.

The orientation attribute is a quaternion describing the rotational component of the transform. The orientation MUST be normalized to have a length of 1.0.

The matrix attribute returns the transform described by the position and orientation attributes as a matrix. This attribute MUST be computed by obtaining the matrix for the XRRigidTransform.

Note: This matrix when premultiplied onto a column vector will rotate the vector by the 3D rotation described by orientation, and then translate it by position. Mathematically in column-vector notation, this is M = T * R, where T is a translation matrix corresponding to position and R is a rotation matrix corresponding to orientation.

The inverse attribute of a XRRigidTransform transform returns an XRRigidTransform in the relevant realm of transform which, if applied to an object that had previously been transformed by transform, would undo the transform and return the object to its initial pose. This attribute SHOULD be lazily evaluated. The XRRigidTransform returned by inverse MUST return transform as its inverse.

An XRRigidTransform with a position of { x: 0, y: 0, z: 0 w: 1 } and an orientation of { x: 0, y: 0, z: 0, w: 1 } is known as an identity transform.

9. Pose

9.1. XRPose

An XRPose describes a position and orientation in space relative to an XRSpace.

[SecureContext, Exposed=Window] interface XRPose {
  [SameObject] readonly attribute XRRigidTransform transform;
  [SameObject] readonly attribute DOMPointReadOnly? linearVelocity;
  [SameObject] readonly attribute DOMPointReadOnly? angularVelocity;

  readonly attribute boolean emulatedPosition;
};

The transform attribute describes the position and orientation relative to the base XRSpace.

The linearVelocity attribute describes the linear velocity in meters per second relative to the base XRSpace. If the user agent can’t populate this, it’s allowed to return null.

The angularVelocity attribute describes the angular velocity in radians per second relative to the base XRSpace. If the user agent can’t populate this, it’s allowed to return null.

The emulatedPosition attribute is false when the transform represents an actively tracked 6DoF pose based on sensor readings, or true if its position value includes a computed offset, such as that provided by a neck or arm model. Estimated floor levels MUST NOT be considered when determining if an XRPose includes a computed offset.

9.2. XRViewerPose

An XRViewerPose is an XRPose describing the state of a viewer of the XR scene as tracked by the XR device. A viewer may represent a tracked piece of hardware, the observed position of a user’s head relative to the hardware, or some other means of computing a series of viewpoints into the XR scene. XRViewerPoses can only be queried relative to an XRReferenceSpace. It provides, in addition to the XRPose values, an array of views which include rigid transforms to indicate the viewpoint and projection matrices. These values should be used by the application when rendering a frame of an XR scene.

[SecureContext, Exposed=Window] interface XRViewerPose : XRPose {
  [SameObject] readonly attribute FrozenArray<XRView> views;
};

The views array is a sequence of XRViews describing the viewpoints of the XR scene, relative to the XRReferenceSpace the XRViewerPose was queried with. Every view of the XR scene in the array must be rendered in order to display correctly on the XR device. Each XRView includes rigid transforms to indicate the viewpoint and projection matrices, and can be used to query XRViewports from layers when needed.

Note: The XRViewerPose’s transform can be used to position graphical representations of the viewer for spectator views of the scene or multi-user interaction.

10. Input

10.1. XRInputSource

An XRInputSource represents an XR input source, which is any input mechanism which allows the user to perform targeted actions in the same virtual space as the viewer. Example XR input sources include, but are not limited to, handheld controllers, optically tracked hands, and gaze-based input methods that operate on the viewer’s pose. Input mechanisms which are not explicitly associated with the XR device, such as traditional gamepads, mice, or keyboards SHOULD NOT be considered XR input sources.

enum XRHandedness {
  "none",
  "left",
  "right"
};

enum XRTargetRayMode {
  "gaze",
  "tracked-pointer",
  "screen",
  "transient-pointer"
};

[SecureContext, Exposed=Window]
interface XRInputSource {
  readonly attribute XRHandedness handedness;
  readonly attribute XRTargetRayMode targetRayMode;
  [SameObject] readonly attribute XRSpace targetRaySpace;
  [SameObject] readonly attribute XRSpace? gripSpace;
  [SameObject] readonly attribute FrozenArray<DOMString> profiles;
  readonly attribute boolean skipRendering;
};

Note: The XRInputSource interface is also extended by the WebXR Gamepads Module

The handedness attribute describes which hand the XR input source is associated with, if any. Input sources with no natural handedness (such as headset-mounted controls) or for which the handedness is not currently known MUST set this attribute "none".

The targetRayMode attribute describes the method used to produce the target ray, and indicates how the application should present the target ray to the user if desired.

• gaze indicates the target ray will originate at the viewer and follow the direction it is facing. (This is commonly referred to as a "gaze input" device in the context of head-mounted displays.)

• tracked-pointer indicates that the target ray originates from either a handheld device or other hand-tracking mechanism and represents that the user is using their hands or the held device for pointing. The orientation of the target ray relative to the tracked object MUST follow platform-specific ergonomics guidelines when available. In the absence of platform-specific guidance, the target ray SHOULD point in the same direction as the user’s index finger if it was outstretched. If the XRSystem determines that part of the handheld device is or becomes intended to contact real-world surfaces (such as a pen tip), the target ray MUST originate at that point.

• screen indicates that the input source was an interaction with the canvas element associated with an inline session’s output context, such as a mouse click or touch event.

• transient-pointer indicates that the input source was generated as part of an operating system interaction intent rather than a specific piece of hardware. Some examples are user intents based on information too sensitive to expose directly such as gaze, synthesised inputs from web driver or inputs generated by assistive technologies. This should only be used for assistive technologies if it is also used as a primary input so as to not inadvertently indicate that assistive technology is being used as per the W3C design principals.

The targetRaySpace attribute is an XRSpace that has a native origin tracking the position and orientation of the preferred pointing ray of the XRInputSource (along its -Z axis), as defined by the targetRayMode.

For input sources with a targetRayMode of "transient-pointer" the targetRaySpace represents the ray to the interaction target at the start of the interaction. The pose for this should be static within the gripSpace for this XRInput.

The gripSpace attribute is an XRSpace that has a native origin tracking to the pose that should be used to render virtual objects such that they appear to be held in the user’s hand. If the user were to hold a straight rod, this XRSpace places the native origin at the centroid of their curled fingers and where the -Z axis points along the length of the rod towards their thumb. The X axis is perpendicular to the back of the hand being described, with the back of the user’s right hand pointing towards +X and the back of the user’s left hand pointing towards -X. The Y axis is implied by the relationship between the X and Z axis, with +Y roughly pointing in the direction of the user’s arm.

The skipRendering attribute indicates that this input is visible and MAY NOT need to be rendered by the current session. If skipRendering is true and the targetRayMode is "tracked-pointer", the user agent MUST ensure that a representation of the XR input source is always shown to the user.

Examples of the controller being shown to the user include the controller is in between the display and the user, the display is transparent or the controller is rendered by the operating system.

skipRendering is a hint to developers about not rendering input sources such as controllers. Pick rays and cursor should still be rendered.

For input sources with a targetRayMode of "transient-pointer" the gripSpace should be the associated user gesture if there is one, otherwise it should be another space the user controls such as the ViewerSpace or the gripSpace or the targetRaySpace of another XRInput. This is to allow user the user to still manipulate the targetRaySpace.

The gripSpace MUST be null if the input source isn’t inherently trackable such as for input sources with a targetRayMode of "gaze" or "screen".

The profiles attribute is a list of input profile names indicating both the prefered visual representation and behavior of the input source.

An input profile name is an ASCII lowercase DOMString containing no spaces, with separate words concatenated with a hyphen (-) character. A descriptive name should be chosen, using the prefered verbiage of the device vendor when possible. If the platform provides an appropriate identifier, such as a USB vendor and product ID, it MAY be used. Values that uniquely identify a single device, such as serial numbers, MUST NOT be used. The input profile name MUST NOT contain an indication of device handedness. If multiple user agents expose the same device, they SHOULD make an effort to report the same input profile name. The WebXR Input Profiles Registry is the recommended location for managing input profile names.

Profiles are given in descending order of specificity. Any input profile names given after the first entry in the list should provide fallback values that represent alternative representations of the device. This may include a more generic or prior version of the device, a more widely recognized device that is sufficiently similar, or a broad description of the device type (such as "generic-trigger-touchpad"). If multiple profiles are given, the layouts they describe must all represent a superset or subset of every other profile in the list.

If the XRSession’s mode is "inline", profiles MUST be an empty list.

The user agent MAY choose to only report an appropriate generic input profile name or an empty list at its discretion. Some scenarios where this would be appropriate are if the input device cannot be reliably identified, no known input profiles match the input device, or the user agent wishes to mask the input device being used.

For example, the Samsung HMD Odyssey’s controller is a design variant of the standard Windows Mixed Reality controller. Both controllers share the same input layout. As a result, the profiles for a Samsung HMD Odyssey controller could be: ["samsung-odyssey", "microsoft-mixed-reality", "generic-trigger-squeeze-touchpad-thumbstick"]. The appearance of the controller is most precisely communicated by the first profile in the list, with the second profile describing an acceptable substitute, and the last profile a generic fallback that describes the device in the roughest sense. (It’s a controller with a trigger, squeeze button, touchpad and thumbstick.)
Similarly, the Valve Index controller is backwards compatible with the HTC Vive controller, but the Index controller has additional buttons and axes. As a result, the profiles for the Valve Index controller could be: ["valve-index", "htc-vive", "generic-trigger-squeeze-touchpad-thumbstick"]. In this case the input layout described by the "valve-index" profile is a superset of the layout described by the "htc-vive" profile. Also, the "valve-index" profile indicates the precise appearance of the controller, while the "htc-vive" controller has a significantly different appearance. In this case the UA would have deemed that difference acceptable. And as in the first example, the last profile is a generic fallback.
(Exact strings are examples only. Actual profile names are managed in the WebXR Input Profiles Registry.)

Note: XRInputSources in an XRSession’s inputSources array are "live". As such, values within them are updated in-place. This means that it doesn’t work to save a reference to an XRInputSource’s attribute on one frame and compare it to the same attribute in a subsequent frame to test for state changes, because they will be the same object. Therefore developers that wish to compare input state from frame to frame should copy the content of the state in question.

An XR input source is a primary input source if it supports a primary action. The primary action is a platform-specific action that, when engaged, produces selectstart, selectend, and select events. Examples of possible primary actions are pressing a trigger, touchpad, or button, speaking a command, or making a hand gesture. If the platform guidelines define a recommended primary input then it should be used as the primary action, otherwise the user agent is free to select one. The device MUST support at least one primary input source.

An XR input source is an tracked input source if it does not support a primary action. These inputs are primarily intended to provide pose data. Note: An example of a tracked input source would be tracking attachments for a users legs or a prop. Tracked hands may also be considered a tracked input source if there is no gesture recognition being performed to detect primary actions.

Each XR input source MAY define a primary squeeze action. The primary squeeze action is a platform-specific action that, when engaged, produces squeezestart, squeezeend, and squeeze events. The primary squeeze action should be used for actions roughly mapping to squeezing or grabbing. Examples of possible primary squeeze actions are pressing a grip trigger or making a grabbing hand gesture. If the platform guidelines define a recommended primary squeeze action then it should be used as the primary squeeze action, otherwise the user agent MAY select one.

Sometimes platform-specific behavior can result in a primary action or primary squeeze action being interrupted or cancelled. For example, an XR input source may be removed from the XR device after the primary action or primary squeeze action is started but before it ends.

10.2. Transient input

Some XR devices may support transient input sources, where the XR input source is only meaningful while performing a transient action, either the primary action for a primary input source, or a device-specific auxiliary action for an tracked input source.

One example would be mouse, touch, or stylus input against an "inline" XRSession, which MUST produce a transient XRInputSource with a targetRayMode set to screen, treated as a primary action for the primary pointer, and as a non-primary auxiliary action for a non-primary pointer.

Another example would be intents from the operating system with input derived from sensitive information that cannot be exposed directly, such as interactions based on gaze. These produce a transient XRInputSource with a targetRayMode set to transient-pointer, treated as a primary action.

Transient input sources are only present in the session’s list of active XR input sources for the duration of the transient action.

Transient input sources follow the following sequence when handling transient actions instead of the algorithms for non-transient primary actions:

10.3. XRInputSourceArray

An XRInputSourceArray represents a list of XRInputSources. It is used in favor of a frozen array type when the contents of the list are expected to change over time, such as with the XRSession inputSources attribute.

[SecureContext, Exposed=Window]
interface XRInputSourceArray {
  iterable<XRInputSource>;
  readonly attribute unsigned long length;
  getter XRInputSource(unsigned long index);
};

The length attribute of XRInputSourceArray indicates how many XRInputSources are contained within the XRInputSourceArray.

The indexed property getter of XRInputSourceArray retrieves the XRInputSource at the provided index.

11. Layers

Note: While this specification only defines the XRWebGLLayer layer, future extensions to the spec are expected to add additional layer types and the image sources that they draw from.

11.1. XRLayer

[SecureContext, Exposed=Window]
interface XRLayer : EventTarget {};

XRLayer is the base class for XRWebGLLayer and other layer types introduced by future extensions.

11.2. XRWebGLLayer

An XRWebGLLayer is a layer which provides a WebGL framebuffer to render into, enabling hardware accelerated rendering of 3D graphics to be presented on the XR device.

typedef (WebGLRenderingContext or
         WebGL2RenderingContext) XRWebGLRenderingContext;

dictionary XRWebGLLayerInit {
  boolean antialias = true;
  boolean depth = true;
  boolean stencil = false;
  boolean alpha = true;
  boolean ignoreDepthValues = false;
  double framebufferScaleFactor = 1.0;
};

[SecureContext, Exposed=Window]
interface XRWebGLLayer: XRLayer {
  constructor(XRSession session,
             XRWebGLRenderingContext context,
             optional XRWebGLLayerInit layerInit = {});
  // Attributes
  readonly attribute boolean antialias;
  readonly attribute boolean ignoreDepthValues;
  attribute float? fixedFoveation;

  [SameObject] readonly attribute WebGLFramebuffer? framebuffer;
  readonly attribute unsigned long framebufferWidth;
  readonly attribute unsigned long framebufferHeight;

  // Methods
  XRViewport? getViewport(XRView view);

  // Static Methods
  static double getNativeFramebufferScaleFactor(XRSession session);
};

Each XRWebGLLayer has a context object, initially null, which is an instance of either a WebGLRenderingContext or a WebGL2RenderingContext.

Each XRWebGLLayer has an associated session, which is the XRSession it was created with.

Note: If an XRWebGLLayer’s composition enabled boolean is set to false all values on the XRWebGLLayerInit object are ignored, since the WebGLRenderingContext’s default framebuffer was already allocated using the context’s actual context parameters and cannot be overridden.

The context attribute is the WebGLRenderingContext the XRWebGLLayer was created with.

Each XRWebGLLayer has a composition enabled boolean which is initially set to true. If set to false it indicates that the XRWebGLLayer MUST NOT allocate its own WebGLFramebuffer, and all properties of the XRWebGLLayer that reflect framebuffer properties MUST instead reflect the properties of the context’s default framebuffer.

The framebuffer attribute of an XRWebGLLayer is an instance of a WebGLFramebuffer which has been marked as opaque if composition enabled is true, and null otherwise. The framebuffer size cannot be adjusted by the developer after the XRWebGLLayer has been created.

An opaque framebuffer functions identically to a standard WebGLFramebuffer with the following changes that make it behave more like the default framebuffer:

• An opaque framebuffer MAY support antialiasing, even in WebGL 1.0.

• An opaque framebuffer’s attachments cannot be inspected or changed. Calling framebufferTexture2D, framebufferRenderbuffer, deleteFramebuffer, or getFramebufferAttachmentParameter with an opaque framebuffer MUST generate an INVALID_OPERATION error.

• An opaque framebuffer has a related session, which is the XRSession it was created for.

• An opaque framebuffer is considered incomplete outside of a requestAnimationFrame() callback. When not in the requestAnimationFrame() callback of its session, calls to checkFramebufferStatus MUST generate a FRAMEBUFFER_UNSUPPORTED error and attempts to clear, draw to, or read from the opaque framebuffer MUST generate an INVALID_FRAMEBUFFER_OPERATION error.

• An opaque framebuffer initialized with depth true will have an attached depth buffer.

• An opaque framebuffer initialized with stencil true will have an attached stencil buffer.

• An opaque framebuffer’s color buffer will have an alpha channel if and only if alpha is true.

• The XR Compositor will assume the opaque framebuffer contains colors with premultiplied alpha. This is true regardless of the premultipliedAlpha value set in the context’s actual context parameters.

Note: User agents are required to respect true values of depth and stencil, which is similar to WebGL’s behavior when creating a drawing buffer.

The buffers attached to an opaque framebuffer MUST be cleared to the values in the table below when first created, or prior to the processing of each XR animation frame. This is identical to the behavior of the WebGL context’s default framebuffer. Opaque framebuffers will always be cleared regardless of the associated WebGL context’s preserveDrawingBuffer value.

Note: Implementations may optimize away the required implicit clear operation of the opaque framebuffer as long as a guarantee can be made that the developer cannot gain access to buffer contents from another process. For instance, if the developer performs an explicit clear then the implicit clear is not needed.

If an XRWebGLLayer is created with an alpha set to true, the framebuffer must be backed by a texture with the RGBA color format. If an XRWebGLLayer is created with an alpha set to false, the framebuffer must be backed by a texture with the RGB color format.

However, the XR Compositor MUST treat the framebuffer’s backing’s pixels as if they were in the SRGB8_ALPHA8 or SRGB8 colorFormat.

NOTE: this means that the XR Compositor MUST not do any gamma conversion from linear RGBA or RGB when it processes the texture backing the framebuffer. Otherwise, the pixels in the final rendering will appear too bright which will not match the rendering on a regular 2D WebGLRenderingContext context.

When an XRWebGLLayer is set as an immersive session’s baseLayer the content of the opaque framebuffer is presented to the immersive XR device immediately after an XR animation frame completes, but only if at least one of the following has occurred since the previous XR animation frame:

• The immersive session’s baseLayer was changed.

• clear, drawArrays, drawElements, or any other rendering operation which similarly affects the framebuffer’s color values has been called while the opaque framebuffer is the currently bound framebuffer of the WebGLRenderingContext associated with the XRWebGLLayer.

Before the opaque framebuffer is presented to the immersive XR device the user agent shall ensure that all rendering operations have been flushed to the opaque framebuffer.

Each XRWebGLLayer has a target framebuffer, which is the framebuffer if composition enabled is true, and the context’s default framebuffer otherwise.

The framebufferWidth and framebufferHeight attributes return the width and height of the target framebuffer’s attachments, respectively.

The antialias attribute is true if the target framebuffer supports antialiasing using a technique of the UA’s choosing, and false if no antialiasing will be performed.

The ignoreDepthValues attribute, if true, indicates the XR Compositor MUST NOT make use of values in the depth buffer attachment when rendering. When the attribute is false it indicates that the content of the depth buffer attachment will be used by the XR Compositor and is expected to be representative of the scene rendered into the layer.

Depth values stored in the buffer are expected to be between 0.0 and 1.0, with 0.0 representing the distance of depthNear and 1.0 representing the distance of depthFar, with intermediate values interpolated linearly. This is the default behavior of WebGL. (See documentation for the depthRange function for additional details.)

Note: Making the scene’s depth buffer available to the compositor allows some platforms to provide quality and comfort improvements such as improved reprojection.

The fixedFoveation attribute controls the amount of foveation used by the XR Compositor. If the user agent or device does not support this attribute, they should return null on getting, and setting should be a no-op. Setting fixedFoveation to a value less than 0 will set it to 0 and setting it to a value higher than 1 will set it to 1. 0 sets the minimum amount of foveation while 1 sets the maximum. It is up to the user agent how the XR Compositor interprets these values. If the fixedFoveation level was changed, it will take effect at the next XRFrame.

NOTE: Fixed foveation is a technique that reduces the resolution content renders at near the edges of the user’s field of view. It can significantly improve experiences that are limited by GPU fill performance. It reduces power consumption and enables applications to increase the resolution of eye textures. It is most useful for low contrast textures, such as background images but less for high contrast ones such as text or detailed images. Authors can adjust the level on a per frame basis to achieve the best tradeoff between performance and visual quality.

Each XRWebGLLayer MUST have a list of full-sized viewports which is a list containing one WebGL viewport for each view the XRSession may expose, including secondary views that are not currently active but may become active for the current session. The viewports MUST have a width and height greater than 0 and MUST describe a rectangle that does not exceed the bounds of the target framebuffer. The viewports MUST NOT be overlapping. If composition enabled is false, the list of full-sized viewports MUST contain a single WebGL viewport that covers the context’s entire default framebuffer.

Each XRWebGLLayer MUST have a list of viewport objects which is a list containing one XRViewport for each active view the XRSession currently exposes.

getViewport() queries the XRViewport the given XRView should use when rendering to the layer.

Each XRSession MUST identify a native WebGL framebuffer resolution, which is the pixel resolution of a WebGL framebuffer required to match the physical pixel resolution of the XR device.

Additionally, the XRSession MUST identify a recommended WebGL framebuffer resolution, which represents a best estimate of the WebGL framebuffer resolution large enough to contain all of the session’s XRViews that provides an average application a good balance between performance and quality. It MAY be smaller than, larger than, or equal to the native WebGL framebuffer resolution. New opaque framebuffer will be created with this resolution, with width and height each scaled by any XRWebGLLayerInit’s framebufferScaleFactor provided.

Note: The user agent is free to use any method of its choosing to estimate the recommended WebGL framebuffer resolution. If there are platform-specific methods for querying a recommended size it is recommended that they be used, but not required. The scale factors used by framebufferScaleFactor and getNativeFramebufferScaleFactor() apply to width and height separately, so a scale factor of two results in four times the overall pixel count. If the platform exposes an area-based render scale that’s based on pixel count, the user agent needs to take the square root of that to convert it to a WebXR scale factor.

11.3. WebGL Context Compatibility

In order for a WebGL context to be used as a source for immersive XR imagery it must be created on a compatible graphics adapter for the immersive XR device. What is considered a compatible graphics adapter is platform dependent, but is understood to mean that the graphics adapter can supply imagery to the immersive XR device without undue latency. If a WebGL context was not already created on the compatible graphics adapter, it typically must be re-created on the adapter in question before it can be used with an XRWebGLLayer.

Note: On an XR platform with a single GPU, it can safely be assumed that the GPU is compatible with the immersive XR devices advertised by the platform, and thus any hardware accelerated WebGL contexts are compatible as well. On PCs with both an integrated and discrete GPU the discrete GPU is often considered the compatible graphics adapter since it generally is a higher performance chip. On desktop PCs with multiple graphics adapters installed, the one with the immersive XR device physically connected to it is likely to be considered the compatible graphics adapter.

Note: "inline" sessions render using the same graphics adapter as canvases, and thus do not need xrCompatible contexts.

partial dictionary WebGLContextAttributes {
    boolean xrCompatible = false;
};

partial interface mixin WebGLRenderingContextBase {
    [NewObject] Promise<undefined> makeXRCompatible();
};

When a user agent implements this specification it MUST set an XR compatible boolean, initially set to false, on every WebGLRenderingContextBase. Once the XR compatible boolean is set to true, the context can be used with layers for any XRSession requested from the current immersive XR device.

Note: This flag introduces slow synchronous behavior and is discouraged. Consider calling makeXRCompatible() instead for an asynchronous solution.

The XR compatible boolean can be set either at context creation time or after context creation, potentially incurring a context loss. To set the XR compatible boolean at context creation time, the xrCompatible context creation attribute must be set to true when requesting a WebGL context. If the requesting document’s origin is not allowed to use the "xr-spatial-tracking" permissions policy, xrCompatible has no effect.

The xrCompatible flag on WebGLContextAttributes, if true, affects context creation by requesting the user agent create the WebGL context using a compatible graphics adapter for the immersive XR device. If the user agent succeeds in this, the created context’s XR compatible boolean will be set to true. To obtain the immersive XR device, ensure an immersive XR device is selected SHOULD be called.

Note: Ensure an immersive XR device is selected needs to be run in parallel, which introduces slow synchronous behavior on the main thread. User agents SHOULD print a warning to the console requesting that makeXRCompatible() be used instead.

function onXRSessionStarted(xrSession) {
  const glCanvas = document.createElement("canvas");
  const gl = glCanvas.getContext("webgl", { xrCompatible: true });

  loadWebGLResources();

  xrSession.updateRenderState({ baseLayer: new XRWebGLLayer(xrSession, gl) });
}

To set the XR compatible boolean after the context has been created, the makeXRCompatible() method is used.

Note: On some systems this flag may turn on a high powered discrete GPU, for example, or proxy all commands to an on-device GPU. If you are in a situation where you may or may not be using XR, it is suggested that you only call makeXRCompatible() when you intend to start an immersive session.

const glCanvas = document.createElement("canvas");
const gl = glCanvas.getContext("webgl");

loadWebGLResources();

glCanvas.addEventListener("webglcontextlost", (event) => {
  // Indicates that the WebGL context can be restored.
  event.canceled = true;
});

glCanvas.addEventListener("webglcontextrestored", (event) => {
  // WebGL resources need to be re-created after a context loss.
  loadWebGLResources();
});

async function onXRSessionStarted(xrSession) {
  // Make sure the canvas context we want to use is compatible with the device.
  // May trigger a context loss.
  await gl.makeXRCompatible();
  xrSession.updateRenderState({ baseLayer: new XRWebGLLayer(xrSession, gl) });
}

12. Events

The task source for all tasks queued in this specification is the XR task source, unless otherwise specified.

12.1. XRSessionEvent

XRSessionEvents are fired to indicate changes to the state of an XRSession.

[SecureContext, Exposed=Window]
interface XRSessionEvent : Event {
  constructor(DOMString type, XRSessionEventInit eventInitDict);
  [SameObject] readonly attribute XRSession session;
};

dictionary XRSessionEventInit : EventInit {
  required XRSession session;
};

The session attribute indicates the XRSession that generated the event.

12.2. XRInputSourceEvent

XRInputSourceEvents are fired to indicate changes to the state of an XRInputSource.

[SecureContext, Exposed=Window]
interface XRInputSourceEvent : Event {
  constructor(DOMString type, XRInputSourceEventInit eventInitDict);
  [SameObject] readonly attribute XRFrame frame;
  [SameObject] readonly attribute XRInputSource inputSource;
};

dictionary XRInputSourceEventInit : EventInit {
  required XRFrame frame;
  required XRInputSource inputSource;
};

The inputSource attribute indicates the XRInputSource that generated this event.

The frame attribute is an XRFrame that corresponds with the time that the event took place. It may represent historical data. getViewerPose() MUST throw an exception when called on frame.

12.3. XRInputSourcesChangeEvent

XRInputSourcesChangeEvents are fired to indicate changes to the list of active XR input sources that are available to an XRSession.

[SecureContext, Exposed=Window]
interface XRInputSourcesChangeEvent : Event {
  constructor(DOMString type, XRInputSourcesChangeEventInit eventInitDict);
  [SameObject] readonly attribute XRSession session;
  [SameObject] readonly attribute FrozenArray<XRInputSource> added;
  [SameObject] readonly attribute FrozenArray<XRInputSource> removed;
};

dictionary XRInputSourcesChangeEventInit : EventInit {
  required XRSession session;
  required sequence<XRInputSource> added;
  required sequence<XRInputSource> removed;

};

The session attribute indicates the XRSession that generated the event.

The added attribute is a list of XRInputSources that were added to the XRSession at the time of the event.

The removed attribute is a list of XRInputSources that were removed from the XRSession at the time of the event.

12.4. XRReferenceSpaceEvent

XRReferenceSpaceEvents are fired to indicate changes to the state of an XRReferenceSpace.

[SecureContext, Exposed=Window]
interface XRReferenceSpaceEvent : Event {
  constructor(DOMString type, XRReferenceSpaceEventInit eventInitDict);
  [SameObject] readonly attribute XRReferenceSpace referenceSpace;
  [SameObject] readonly attribute XRRigidTransform? transform;
};

dictionary XRReferenceSpaceEventInit : EventInit {
  required XRReferenceSpace referenceSpace;
  XRRigidTransform? transform = null;
};