Navigation component

This component defines nodes to control the camera, and some effects closely related to the camera. NavigationInfo controls the camera behavior. Viewpoint and OrthoViewpoint define the initial camera position and rotation, and may be used to animate the camera. LOD allows to implement level-of-detail, where different versions of some object are displayed depending on the camera distance. Collision allows to use a simpler geometry for collision purposes, or even to turn the collisions off (like a fake wall hiding a secret room in games). Billboard may be used to create sprites, as it aligns geometry (flat or not) with respect to the camera.

See also X3D specification of the Navigation component.

Contents:

See also Castle Game Engine (and view3dscene) extensions related to navigation.

1. Demos

For demos and tests of these features, see the navigation subdirectory inside our VRML/X3D demo models.

2. Supported nodes

  • Viewpoint(Pascal API: TViewpointNode),
    OrthoViewpoint(Pascal API: TOrthoViewpointNode).

    These nodes define the initial camera position, rotation, and field of view. The Viewpoint makes a perspective projection (objects further away are smaller), OrthoViewpoint makes an orthographic projection.

    You can even animate their transformation and properties, to animate the camera.

    Note: In Castle Game Engine you can also control the camera by Pascal code, using the Viewport.Camera property that keeps a TCastleCamera instance. See manual about controlling the camera. Still, this X3D node is useful to set initial camera through e.g. Blender (you can export to glTF or X3D and importing them will set initial camera properly).

    view3dscene displays a nice menu allowing you to jump to any defined viewpoint, showing viewpoints' descriptions. Animating viewpoint's position and orientation (directly or by animating it's parent transformation) also works perfectly.

    Note that merely adding a Viewpoint (or OrthoViewpoint) node to the X3D model, with everything at default, changes how view3dscene and CGE interpret the model.

    • Without the Viewpoint (or OrthoViewpoint), we determine center of rotation (important for proper "Examine" work) based on scene contents (bounding box middle).

    • With Viewpoint (or OrthoViewpoint), we assume you have set proper centerOfRotation. The default centerOfRotation is just (0,0,0). This follows X3D standard.

      You can set Viewport.autoCenterOfRotation (or OrthoViewport.autoCenterOfRotation) to true to explicitly request to automatically determine (based on bounding box middle) this center of rotation. This field is a CGE extension.

    TODO: We support most, but not all, X3D fields. Fields not implemented yet: jump, retainUserOffsets.

    • As for animating viewpoints (and also possible jump, retainUserOffsets implementation):

      We do not support right now the notion of preserving user offsets from navigation. Tracking them in the past required to have a special treatment of camera transformation (initial vs current), but it was too complicated too keep (as we wanted to make TCastleCamera descend from TCastleTransform) and it didn't seem useful for authors in reality. (Of course please report if it is useful for you!)

      When you animate the viewpoint (either by animating viewpoint position/orientation or by animating viewpoint transformation) right now we just change the current camera to match the viewpoint. User offsets are not retained.

  • NavigationInfo(Pascal API: TNavigationInfoNode)

    Controls the navigation behavior (mode of movement, gravity, etc.) and the headlight.

    Note: In Castle Game Engine you can also control the navigation by Pascal code. See manual about navigation. And you can control the headlight by just adding lights as camera children. Still, this node is useful to define defaults.

    Details about supported fields:

    • avatarSize is honoured fully:

      1. First avatarSize item is the camera radius. If you use view3dscene, note that --camera-radius command-line option overrides this value.

      2. 2nd avatarSize item is the preferred height above the ground.

      3. 3rd avatarSize item is the tallest object over which you can climb.

        If this is missing (or it has value <= 0) then there's no such limit, and you can climb as long as you can move forward. So you can climb the stairs with steps almost as high as your own height minus the camera radius. Simplifying (ignoring other effects, like head bobbing), you can say that avatarSize[2] is by default like avatarSize[1] - avatarSize[0]).

        See TCastleWalkNavigation.ClimbHeight API docs for more details about this.

    • speed field is supported, it sets the speed in meters/second. Speed = 0.0 is also correctly honored (user will not be able to move in Walk/Fly modes, only to rotate).

    • type field is supported. Navigation types fully supported are: EXAMINE, WALK, FLY, NONE. Inside the engine, the navigation paradigm is actually a little more flexible. You can fine-tune the rotations and gravity behavior by view3dscene Navigation -> Walk and Fly Settings menu.

      As an extension, we also support new navigation mode TURNTABLE. This is similar to the Examine mode, with controls comfortable for viewing models that have a sense of floor/ground in the XZ plane, and vertical axis in +Y. Implementation is not finished yet.

      The presence of navigation type ANY is ignored by view3dscene. We always show controls to change navigation settings, hiding them feels harmful to user.

    • Nice transitions between viewpoints are supported, honouring transitionType and transitionTime fields, and (since view3dscene 3.13.0) making transitionComplete event. See demo model navigation/transition_multiple_viewpoints.x3dv showing how to use it to make an animated transition between a couple of viewpoints.

    Binding different NavigationInfo nodes, and changing their exposed fields by events, of course works.

    When no NavigationInfo node is present in the scene:

    • avatarSize and speed are calculated based on scene's bounding box sizes, to a values that will hopefully "feel right". We try to calculate them intelligently, because simply using NavigationInfo defaults results in bad experience in many scenes.
    • headlight behaves like true, type behaves like [EXAMINE, ANY], this follows NavigationInfo defaults.

    TODO: visibilityLimit may be ignored if shadow volumes are allowed (We use frustum with z-far in infinity then.)

  • LOD(Pascal API: TLODNode)

    Allows to define various versions of the same object, with varying level-of-detail. An appropriate child is automatically displayed based on the current distance to the camera.

    Note: Right now, the only thing that decides "which level of detail should be used" is the distance to the camera. Which means that only the supplied LOD.range controls which level is displayed. The forceTransitions value is simply ignored, and when range is empty, we simply always use the first (highest-detail) version.

    Example: The simplest example is part of our demo models. View the X3D code here: navigation/lod_test.x3dv.

  • Billboard(Pascal API: TBillboardNode)

    The children of this node are automatically aligned with the camera. Two modes are possible: where the objects are rotated only around a specified axis, or where objects are rotated freely to match the camera.

    Example: The simplest example is part of our demo models. View the X3D code here: navigation/billboard_simple.x3dv.

  • Collision(Pascal API: TCollisionNode)

    Allows to define a simpler geometry for collision purposes, or even to turn the collisions off (like a fake wall hiding a secret room in games).

    Most things work: grouping (children property, in particular), allows to control collision detection by honoring enabled (named collide in VRML 97) and proxy fields.

    bboxCenter/Size is currently simply ignored, our engine always calculates and updates the bounding boxes where needed.

    TODO: collideTime and isActive out events are not implemented yet.

  • ViewpointGroup(Pascal API: TViewpointGroupNode)

    You can use this to create submenus in "Viewpoints" menu in view3dscene. Fields description and children are taken into account. Also, you can use this node to hide some viewpoints from the menu: the displayed field also works.

    TODO: size/center is not honored yet. Group is displayed regardless of camera position. A possible workarond could be to use a ProximitySensor node, routing ProximitySensor.isActive to the displayed field... Except this workaround will not work too, because changing of the displayed field after the scene loading doesn't change the menu for now.

    TODO: retainUserOffsets is ignored.