Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
1 contributor

Users who have contributed to this file

301 lines (230 sloc) 26.2 KB

WebVR to WebXR Migration Guide

This guide is intended to help developers that have created content with the now deprecated WebVR API migrate to the WebXR API that replaced it. For simplicity this document is going to primarily focus on the differences between how WebVR and WebXR presenting VR content to a headset, since that’s the primary draw of the API. Differences in displaying inline content are not covered.

Secure origin required

The WebXR API is considered a "powerful feature" and thus only available on secure origins (ie: URLs using HTTPS). For development purposes localhost counts as a secure origin, and other domains can be temporarily treated as secure via browser-specific mechanisms:

Hardware enumeration

WebVR applications start by calling navigator.getVRDisplays(), which returns a list of connected VR hardware. The developer then chooses a VRDisplay from the list and keeps a reference to it. This object is what almost all further interaction is done with for the remainder of the application. Changes to the available set of VR hardware is indicated with the vrdisplayconnect and vrdisplaydisconnect events on the Window object.

In WebXR a list of connected hardware cannot be retrieved to avoid fingerprinting. Instead a single active “XR device” is implicitly picked by the UA and all operations are performed against it. (System support for multiple XR devices at once is almost unheard of, so this isn’t problematic for most any real world scenario.) Changes to the available VR hardware are indicated by the devicechange event of the navigator.xr object.

Testing for support

Both WebVR and WebXR applications have a need to advertise to users that VR content can be shown, usually by enabling a button that will be used to initiate the VR content’s presentation to the display.

WebVR applications test to see if the VRDisplay will allow VR content to be shown on it by checking the vrDisplay.capabilities.canPresent boolean attribute.

WebXR applications call navigator.xr.isSessionSupported() with the desired XRSessionMode (which indicates the type of content they want to display, such as “immersive-vr”) and check the resolved boolean to see if the implicit XR Device supports it.

WebVR

// All code samples will omit error checking for clarity.
let displays = await navigator.getVRDisplays();
let vrDisplay = displays[0];
if (vrDisplay.capabilities.canPresent) {
  ShowEnterVRButton();
}

WebXR

let supported = await navigator.xr.isSessionSupported('immersive-vr');
if (supported) {
  ShowEnterVRButton();
}
});

Starting VR presentation

WebVR applications begin presenting VR content by calling vrDisplay.requestPresent().

WebXR applications begin presenting VR content by calling navigator.xr.requestSession() with the XRSessionMode of “immersive-vr”. This returns a promise that resolves to an XRSession, which the developer keeps a reference to. This object is what almost all further interaction is done with for the remainder of the VR content presentation.

Both methods must be called during a user activation event.

Rendering Setup

Both APIs render VR content using WebGL, though the way that imagery is supplied to the API differs.

WebVR applications pass an HTMLCanvasElement, expected to have an attached WebGLRenderingContext. as the source of a VRLayerInit dictionary, which is then passed to vrDisplay.requestPresent(). From then on any content rendered to the canvas’ WebGL context’s backbuffer is presented to the VR hardware. For best results, it is expected that the application will resize the canvas to match the combined renderWidth and renderHeight reported by the VREyeParameters for both eyes, reported by calling vrDisplay.getEyeParameters() with the desired VREye.

WebXR applications must first make the WebGL context compatible with the active XR device. This ensures that the WebGL resources reside on the GPU that is optimal for VR rendering (for example, the one that the headset is physically connected to on a multi-GPU desktop PC). This is done by either setting the xrCompatible key to true in the WebGLContextCreationAttributes when creating the context or calling gl.makeXRCompatible() on the context after it’s been created (which may trigger a context loss). Then the developer constructs a new XRWebGLLayer, passing in both the XRSession and an XR compatible WebGLRenderingContext. This layer is then set as the source of the content the VR hardware will display by passing it to xrSession.updateRenderState() as the baseLayer of the XRRenderStateInit dictionary. The canvas does not need to be resized for best results.

WebVR

let glCanvas = document.createElement('canvas');
let gl = document.getContext('webgl');

await vrDisplay.requestPresent([{ source: glCanvas }]);
let leftEye = vrDisplay.getEyeParameters("left");
let rightEye = vrDisplay.getEyeParameters("right");

glCanvas.width = Math.max(leftEye.renderWidth, rightEye.renderWidth) * 2;
glCanvas.height = Math.max(leftEye.renderHeight, rightEye.renderHeight);

// Now presenting to the headset.

WebXR

let glCanvas = document.createElement('canvas');
let gl = document.getContext('webgl', { xrCompatible: true });

let xrSession = await navigator.xr.requestSession('immersive-vr');
let xrLayer = new XRWebGLLayer(session, gl);
session.updateRenderState({ baseLayer: xrLayer });

// Now presenting to the headset.

Tracking Setup

WebVR has a single implicit tracking environment that all poses are delivered in. In order to allow the user to feel like the floor of their virtual environment aligns with the floor of their physical environment they must transform all poses by the sittingToStandingTransform matrix that’s reported by the vrDisplay.stageParameters attribute. If the developers wants to know the boundaries of the user’s play space they can look at the sizeX and sizeZ attributes of the vrDisplay.stageParameters, which describe an axis aligned rectangle centered on the origin defined by the sittingToStandingTransform. (This limited form of boundaries reporting may force the reported size to be significantly smaller than the actual boundaries the user configured.)

WebXR requires the developer to define the tracking environment they want poses communicated in. This is both to enable a wider range of hardware (like AR devices) and to simplify the creation of floor-aligned content by removing much of the matrix math that WebVR required. The tracking space is specified by calling xrSession.requestReferenceSpace() with the desired XRReferenceSpaceType, which returns a promise that resolves to an XRReferenceSpace. The developer will supply this object any time poses are requested. A “local” reference space closely aligns with WebVR’s implicit tracking environment, while a “local-floor” reference space aligns the virtual environment with the floor of the user’s physical environment similar to WebVR’s sittingToStandingTransform (but with less math expected of the developer in order to make it work.) A “bounded-floor” reference space also aligns with the user’s physical floor, with the addition of reporting boundsGeometry, which gives a full polygonal boundary that’s more flexible/accurate than WebVR’s rectangular equivalent.

WebVR

// No equivalent

WebXR

xrReferenceSpace = await xrSession.requestReferenceSpace("local");

If the developer wants to use reference spaces other than "local" during an ["immersive-vr"] session they must also request consent to use it at session creation time by passing the desired type to either the requiredFeatures or optionalFeatures members of the XRSessionInit dictionary passed to navigator.xr.requestSession(). This will cause the UA to prompt the user for their consent to use the more detailed levels of tracking if necessary.

WebVR

// No equivalent

WebXR

let xrSession = await navigator.xr.requestSession('immersive-vr', {
  requiredFeatures: ["local-floor"]
});

let xrReferenceSpace = await xrSession.requestReferenceSpace("local-floor");

Animation Loop

Both VRDisplay and XRSession have a requestAnimationFrame function that is called to process rendering callbacks at the appropriate refresh rate for the VR hardware.

In WebVR during the vrDisplay.requestAnimationFrame() callback the user’s pose is queried by calling vrDisplay.getFrameData(), which is passed an application-allocated VRFrameData object to populate with the current pose data. The VRFrameData contains projection and view matrices for the user’s left and right eyes, as well as a VRPose that describes the position, orientation, velocity, and acceleration of the user at the time of the frame. The application is expected to use the projection and view matrices as-is, even though it may appear they could be computed from the VRPose and values given in the VREyeParameters.

In WebXR an XRFrame is passed into the callback provided to xrSession.requestAnimationFrame(). The user’s pose is queried from the XRFrame by calling xrFrame.getViewerPose() with the XRReferenceSpace the developer wants the pose reported in. The XRViewerPose that’s returned contains an array of XRViews, each of which reports a projectionMatrix and a transform that indicates the required position of the “camera” for that view. The projectionMatrix is expected to be used as-is, but the transform (which is an XRRigidTransform) providing a position vector and orientation quaternion, as well as a matrix representation of the same transform.) The XRViewerPose also has a top-level transform that gives the position and orientation for the VR hardware. No velocity or acceleration is exposed by WebXR at this time.

In WebVR the application is always expected to render the scene twice, once for the left eye to the left half of the default WebGL framebuffer, and once for the right eye to the right half of the default WebGL framebuffer. The resolution that the app renders at can be controlled two ways: By resizing the WebGL backbuffer (with the canvas width and height attributes) or by changing the left and right viewports by setting the leftBounds and rightBounds of the VRLayerInit when calling vrDisplay.requestPresent().

In WebXR the application renders the scene N times, once for each XRView that’s reported by the XRViewerPose. The number of views reported may change from frame to frame. Content is rendered into the framebuffer of the XRWebGLLayer, which is allocated by the UA to match the VR hardware's needs. (The default WebGL framebuffer is not used by WebXR for "immersive-vr" sessions, and can be rendered into for display on the page as usual during VR presentation.) The viewport for each view is determined by passing the XRView into xrWebGLLayer.getViewport(). The size of the XRWebGLLayer framebuffer is determined by the VR hardware (and reported on the layer as framebufferWidth and framebufferHeight) but can be scaled at layer creation time by setting the framebufferScaleFactor in the XRWebGLLayerInit dictionary.

When a WebVR application is done rendering it must call vrDevice.submitFrame() to capture the content rendered to the WebGL context’s default framebuffer and display it on the VR hardware.

WebXR automatically presents the content of the XRWebGLLayer’s framebuffer to the VR hardware when the xrSession.requestAnimationFrame() callback returns.

WebVR

let vrFrameData = new VRFrameData();
function onFrame (t) {
  // Queue a request for the next frame to keep the animation loop going.
  vrDisplay.requestAnimationFrame(onFrame);

  // Get the frame data, which contains the required matrices.
  vrDisplay.getFrameData(frameData);

  // Ensure we're rendering to the default backbuffer.
  gl.bindFramebuffer(gl.FRAMEBUFFER, null);

  // Set the default left eye viewport, assuming that it wasn't changed during 
  // the call to vrDisplay.requestPresent().
  gl.viewport(0, 0, webglCanvas.width * 0.5, webglCanvas.height);

  // Render the scene using a fictional rendering library with the left eye's
  // projection and view matrix.
  scene.setProjectionMatrix(frameData.leftProjectionMatrix);
  scene.setViewMatrix(frameData.leftViewMatrix);
  scene.render();

  // Set the default right eye viewport.
  gl.viewport(webglCanvas.width * 0.5, 0, webglCanvas.width * 0.5, webglCanvas.height);

  // Render the scene using a fictional rendering library with the right eye's
  // projection and view matrix.
  scene.setProjectionMatrix(frameData.rightProjectionMatrix);
  scene.setViewMatrix(frameData.rightViewMatrix);
  scene.render();

  vrDisplay.submitFrame();
}

WebXR

function onFrame(t, frame) {
  let session = frame.session;
  // Queue a request for the next frame to keep the animation loop going.
  session.requestAnimationFrame(onXRFrame);

  // Get the XRDevice pose relative to the Reference Space we created
  // earlier. The pose may not be available for a variety of reasons, so
  // we'll exit the callback early if it comes back as null.
  let pose = frame.getViewerPose(xrReferenceSpace);
  if (!pose) {
    return;
  }

  // Ensure we're rendering to the layer's backbuffer.
  let layer = session.renederState.baseLayer;
  gl.bindFramebuffer(gl.FRAMEBUFFER, layer.framebuffer);

  // Loop through each of the views reported by the viewer pose.
  for (let view of pose.views) {
    // Set the viewport required by this view.
    let viewport = layer.getViewport(view);
    gl.viewport(viewport.x, viewport.y, viewport.width, viewport.height);

    // Render the scene using a fictional rendering library with the view's
    // projection matrix and view transform.
    scene.setProjectionMatrix(view.projectionMatrix);
    scene.setCameraTransform(view.transform.position, view.transform.orientation);
    // Alternatively, the view matrix can be retrieved directly like so:
    // scene.setViewMatrix(view.transform.inverse.matrix);
    scene.render(view.projectionMatrix, view.transform);
  }
}

Input

WebVR applications receive VR controller input from the navigator.getGamepads() API. Some of the Gamepad objects returned will have a non-zero displayId, which indicates they are associated with a VRDisplay. The buttons and axes arrays indicate input states as usual, while a non-standard pose attribute on the Gamepad indicates the VRPose of the controller in WebVR’s implicit tracking space. If the developer wants the controller’s pose relative to the user’s floor it must be manually transformed with the vrDisplay.stageParameters.sittingToStandingTransform. A non-standard hand attribute on the Gamepad indicates which hand the controller is associated with, if known. The id attribute of the Gamepad indicates the name of the controller, and can be used to load an appropriate mesh to represent the device in the virtual scene.

WebXR applications surface input from multiple sources through xrSession.inputSources, which is an array of XRInputSource objects. The XRInputSource contains a targetRaySpace (for point and click tracking) and optional gripSpace (for handheld objects) which can be passed to the xrFrame.getPose() method along with the tracking space they should be reported relative to in order to get the XRPose of the input source. The XRInputSource also has a handedness attribute to indicate which hand the input source is associated with, if known. For button and axis state the XRInputSource has an optional gamepad attribute, which is a Gamepad object (that notably lacks the non-standard extensions used by WebVR and does not appear in the array returned by navigator.getGamepads()). The profiles attribute of the XRInputSource contains an array of strings that indicate, with decreasing specificity, the type of input device and can be used to load an appropriate mesh to represent the device in the virtual scene.

It is not possible to trigger user activation events with WebVR input.

The selectstart, select, and selectend events fired on an XRSession indicate when the primary trigger, button, or gesture of an XRInputSource is being interacted with, and can be used to facilitate basic interaction without the need to observe the gamepad state. The select event is a user activation event and can be used to begin media playback, among other things.

There is also a corresponding set of squeezestart, squeeze, and squeezeend events that are fired when either a grip button or squeeze gesture is being interacted with. The squeeze event also is a user activation event.

WebVR

function onFrame (t) {
  // Queue a request for the next frame to keep the animation loop going.
  vrDisplay.requestAnimationFrame(onFrame);

  // Loop through all gamepads and identify the ones that are associated with
  // the vrDisplay.
  let gamepads = navigator.getGamepads();
  for (let i = 0; i < gamepads.length; ++i) {
    let gamepad = gamepads[i];
    // The array may contain undefined gamepads, so check for that as
    // well as a non-null pose.
    if (gamepad && gamepad.displayId && gamepad.pose) {
      scene.showControllerAtTransform(gamepad.pose.position, gamepad.pose.orientation, gamepad.hand);
    }
  }

  // Handle rendering as shown above...
}

WebXR

function onFrame(t, frame) {
  // Queue a request for the next frame to keep the animation loop going.
  xrSession.requestAnimationFrame(onXRFrame);

  // Loop through all input sources.
  for (let inputSource of xrSession.inputSources) {
    // Show the input source if it has a grip space
    if (inputSource.gripSpace) {
      let inputPose = frame.getPose(inputSource.gripSpace, xrReferenceSpace);
      scene.showControllerAtTransform(inputPose.position, inputPose.orientation, inputSource.handedness);
    }
  }

  // Handle rendering as shown above...
);

Ending VR presentation

Both WebVR and WebXR may have their presentation of VR content ended by the UA at any time.

WebVR applications may explicitly end the presentation of VR content by calling vrDisplay.exitPresent(). A vrdisplaypresentchange event is fired on the Window object when presentation is started or ended by either the application or the UA. The presentation state is determined by checking the vrDisplay.isPresenting boolean attribute.

WebXR applications may explicitly end the presentation of VR content by calling xrSession.end(), at which point the XRSession object becomes unusable. An end event is fired on the XRSession when it is ended by either the application or UA.

WebVR

window.addEventListener("vrdisplaypresentchange", () => {
  if (!vrDisplay.isPresenting) {
    // VR presentation has ended. Do any necessary cleanup.
  }
});

WebXR

xrSession.addEventListener("end", () => {
  // VR presentation has ended. Do any necessary cleanup.
});

Misc

WebXR currently has no equivalent for the vrdisplayactivate, vrdisplaydeactivate, vrdisplaypointerrestricted, and vrdisplaypointerunrestricted events. WebXR events also have no equivalent of the reason enum reported for VRDisplayEvent types.

WebXR currently has no equivalent for the VRDisplayCapabilities.hasExternalDisplay attribute.

WebXR has no method for reporting projection parameters in terms of field of view.

Additional Resources

Developers looking for more information about how to use WebXR can also refer to the following resources:

You can’t perform that action at this time.