Scene Graph Superstructure

Java 3D's superstructure consists of one or more VirtualUniverse objects, each of which contains a set of one or more high-resolution Locale objects. The Locale objects, in turn, contain collections of subgraphs that comprise the scene graph (see Figure 1).

The Virtual Universe

Java 3D defines the concept of a virtual universe as a three-dimensional space with an associated set of objects. Virtual universes serve as the largest unit of aggregate representation, and can also be thought of as databases. Virtual universes can be very large, both in physical space units and in content. Indeed, in most cases a single virtual universe will serve an application's entire needs.

Virtual universes are separate entities in that no node object may exist in more than one virtual universe at any one time. Likewise, the objects in one virtual universe are not visible in, nor do they interact with objects in, any other virtual universe.

To support large virtual universes, Java 3D introduces the concept of Locales that have high-resolution coordinates as an origin. Think of high-resolution coordinates as "tie-downs" that precisely anchor the locations of objects specified using less precise floating-point coordinates that are within the range of influence of the high-resolution coordinates.

A Locale, with its associated high-resolution coordinates, serves as the next level of representation down from a virtual universe. All virtual universes contain one or more high-resolution-coordinate Locales, and all other objects are attached to a Locale. High-resolution coordinates act as an upper-level translation-only transform node. For example, the coordinates of all objects that are attached to a particular Locale are all relative to the location of that Locale's high-resolution coordinates.

The Virtual Universe

While a virtual universe is similar to the traditional computer graphics concept of a scene graph, a given virtual universe can become so large that it is often better to think of a scene graph as the descendant of a high-resolution-coordinate Locale.

Establishing a Scene

To construct a three-dimensional scene, the programmer must execute a Java 3D program. The Java 3D application must first create a VirtualUniverse object and attach at least one Locale to it. Then the desired scene graph is constructed, starting with a BranchGroup node and including at least one ViewPlatform object, and the scene graph is attached to the Locale. Finally, a View object that references the ViewPlatform object (see "Structuring the Java 3D Program") is constructed. As soon as a scene graph containing a ViewPlatform is attached to the VirtualUniverse, Java 3D's rendering loop is engaged, and the scene will appear on the drawing canvas(es) associated with the View object.

Loading a Virtual Universe

Java 3D is a runtime application programming interface (API), not a file format. As an API, Java 3D provides no direct mechanism for loading or storing a virtual universe. Constructing a scene graph involves the execution of a Java 3D program. However, loaders to convert a number of standard 3D file formats to or from Java 3D virtual universes are expected to be generally available.

Coordinate Systems

By default, Java 3D coordinate systems are right-handed, with the orientation semantics being that +y is the local gravitational up, +x is horizontal to the right, and +z is directly toward the viewer. The default units are meters.

High-Resolution Coordinates

Double-precision floating-point, single-precision floating-point, or even fixed-point representations of three-dimensional coordinates are sufficient to represent and display rich 3D scenes. Unfortunately, scenes are not worlds, let alone universes. If one ventures even a hundred miles away from the (0.0, 0.0, 0.0) origin using only single-precision floating-point coordinates, representable points become quite quantized, to at very best a third of an inch (and much more coarsely than that in practice).

To "shrink" down to a small size (say the size of an IC transistor), even very near (0.0, 0.0, 0.0), the same problem arises.

If a large contiguous virtual universe is to be supported, some form of higher-resolution addressing is required. Thus the choice of 256-bit positional components for "high-resolution" positions.

Java 3D High-Resolution Coordinates

Java 3D high-resolution coordinates consist of three 256-bit fixed-point numbers, one each for x, y, and z. The fixed point is at bit 128, and the value 1.0 is defined to be exactly 1 meter. This coordinate system is sufficient to describe a universe in excess of several hundred billion light years across, yet still define objects smaller than a proton (down to below the planck length). Table 1 shows how many bits are needed above or below the fixed point to represent the range of interesting physical dimensions.

Table 1 – Java 3D High-Resolution Coordinates
2n Meters Units
87.29 Universe (20 billion light years) 
69.68 Galaxy (100,000 light years)
53.07 Light year
43.43 Solar system diameter
23.60 Earth diameter
10.65 Mile
9.97 Kilometer
0.00 Meter
-19.93 Micron
-33.22 Angstrom
-115.57 Planck length

A 256-bit fixed-point number also has the advantage of being able to directly represent nearly any reasonable single-precision floating-point value exactly.

High-resolution coordinates in Java 3D are used only to embed more traditional floating point coordinate systems within a much higher-resolution substrate. In this way a visually seamless virtual universe of any conceivable size or scale can be created, without worry about numerical accuracy.

Java 3D Virtual World Coordinates

Within a given virtual world coordinate system, positions are expressed by three floating point numbers. The virtual world coordinate scale is in meters, but this can be affected by scale changes in the object hierarchy.

Details of High-Resolution Coordinates

High-resolution coordinates are represented as signed, two's-complement, fixed-point numbers consisting of 256 bits. Although Java 3D keeps the internal representation of high-resolution coordinates opaque, users specify such coordinates using 8-element integer arrays. Java 3D treats the integer found at index 0 as containing the most significant bits and the integer found at index 7 as containing the least significant bits of the high-resolution coordinate. The binary point is located at bit position 128, or between the integers at index 3 and 4. A high-resolution coordinate of 1.0 is 1 meter.

The semantics of how file loaders deal with high-resolution coordinates is up to the individual file loader, as Java 3D does not directly define any file-loading semantics. However, some general advice can be given (note that this advice is not officially part of the Java 3D specification).

For "small" virtual universes (on the order of hundreds of meters across in relative scale), a single Locale with high-resolution coordinates at location (0.0, 0.0, 0.0) as the root node (below the VirtualUniverse object) is sufficient; a loader can automatically construct this node during the loading process, and the point in high-resolution coordinates does not need any direct representation in the external file.

Larger virtual universes are expected to be constructed usually like computer directory hierarchies, that is, as a "root" virtual universe containing mostly external file references to embedded virtual universes. In this case, the file reference object (user-specific data hung off a Java 3D group or hi-res node) defines the location for the data to be read into the current virtual universe.

The data file's contents should be parented to the file object node while being read, thus inheriting the high-resolution coordinates of the file object as the new relative virtual universe origin of the embedded scene graph. If this scene graph itself contains high-resolution coordinates, it will need to be offset (translated) by the amount in the file object's high-resolution coordinates and then added to the larger virtual universe as new high-resolution coordinates, with their contents hung off below them. Once again, this procedure is not part of the official Java 3D specification, but some more details on the care and use of high-resolution coordinates in external file formats will probably be available as a Java 3D application note.

Authoring tools that directly support high-resolution coordinates should create additional high-resolution coordinates as a user creates new geometry "sufficiently" far away (or of different scale) from existing high-resolution coordinates.

Semantics of widely moving objects. Most fixed and nearly-fixed objects stay attached to the same high-resolution Locale. Objects that make wide changes in position or scale may periodically need to be reparented to a more appropriate high-resolution Locale. If no appropriate high-resolution Locale exists, the application may need to create a new one.

Semantics of viewing. The ViewPlatform object and the associated nodes in its hierarchy are very often widely moving objects. Applications will typically attach the view platform to the most appropriate high-resolution Locale. For display, all objects will first have their positions translated by the difference between the location of their high-resolution Locale and the view platform's high-resolution Locale. (In the common case of the Locales being the same, no translation is necessary.)