VRML The potential of VRML for visualising space and interactions within it.


This document discusses potential uses for VRML (Virtual Reality Modelling Language) scenarios. The majority of the projects were carried out during a guest research position at the Space and Virtuality Studio of the Interactive Institute. In addition to more general VRML concepts a number of projects present possible virtual solutions to complex spatial and interaction scenarios encountered in some of the research themes within the studio.


The specifications for VRML or the Virtual Reality Modeling Language were first published by Gavin Bell, Anthony Parisi and Mark Pesce in November 1995. VRML aimed to describe 3D spaces or 'worlds' in a similar way to HTML, i.e. in a file format which works across computer platforms and consists of file sizes small enough for fast download times.

forest.wrl: 16KB
One of the first 'worlds' that I encountered was a poetic model of a forest, created by Peter Hughes and was available throughout 1996 at http://www.cruz.com/~hughes/. The model on the left has been converted to the VRML 2 specifications which were published late in 1997 by 'The VRML Consortium Incorporated'. Since then VRML models turned up in all fields of research from molecular chemistry to metereology and from medicine to statistics. Two main application areas today are in the fields of architecture and in web communication where VRML is often used to provide basic model for virtual communities.

By the time I encountered VRML late in 1995 I had extensive experience with 3D modelling both in architectural CAD software and in high-end computer animation software. Neither allowed for navigating the models or worlds in shaded mode, i.e. whenever the camera position was changed, the window would redraw in wireframe mode first. With VRML it was possible for the first time to navigate relatively smoothly through a world which, although jerky at times, stayed 'solid' and thus enabled the viewer to get some continuous sense of immersion.

DKG.wrl: 74KB

Thus VRML led to a change in my work practice. No longer did I produce time-consuming rendered still images and video sequences to evaluate my designs. Instead they were converted to VRML where I could get a feeling for the curvature and distances of a design before constructing them as full-scale models in galleries. The 'DKG.wrl' file on the left is an example from my art practice converting an existing gallery space into an amorphous environment. Within the VRML file the existing building is represented through red outlines only whereas the amorphous design is shown in white representing the corrugated cardboard used for the actual full-scale conversion.

Morph.wrl: 143KB

The 'Morph.wrl' example shows a complex morph in 3 dimensions changing a cube into a head consisting of approximately 800 vertex points. The geometric models were created by utilising the Haptic Workbench at the CSIRO in Canberra, Australia. A gradual change in colour is possible by interpolating the 'colorPerVertex' node.

HOST.wrl: 24KB
soundfile 1: 30KB
soundfile 2: 9KB

The final example here, HOST.wrl, aimed at visualising complex choreographies between a moveable architectural set and theatrical performers in preparation for a collaboration with the performance group Gravity Feed. In this model the camera viewpoint and various geometrical objects are animated, the browser interface is hidden during the animation and simple sound effects are triggered towards the end. Originally I hoped that a large part of the performance could be visualised in this way before the event. Unfortunately there was not enough time to set up a system that would allow for changes as fast as the performers came up with ideas.


How to create VRML models

CADfile.wrl: 40KB

The most obvious choice would be 3D modelling software and during the last few years VRML export tools have become standard. Unfortunately the majority of them do not consider the special requirements for efficient virtual reality display on the one hand and for World Wide Web usage on the other. Designs in virtual reality function efficiently only if the polygon count is low, if there are few light sources and if shape hints such as convex polygons, solid primitives, backface culling etc. are explicitly stated.

World Wide Web usage requires small file sizes which can be achieved through instancing, i.e. defining and then referencing modular units. The 'CADfile.wrl' on the left is an example from a commercial software package which converted the cylinder into an 'indexFaceSet' instead of a shape primitive. This not only created an incorrect, angular look but also a long list of coordinate points. In addition the program exported every coordinate point of the 'hooks' instead of writing out one set and instancing it along the cylinder. The resulting file size of 40 Kb is unnecessarily large.

Detail of VRML file with a cylinder unnecessarily converted to polygons.

FS2000.wrl: 2KB

In contrast to the 'CADfile.wrl' above, the file on the left (FS2000.wrl) has been created entirely in a word editor by making full use of the VRML 2 Specifications. Despite added details such as the endplates and a connecting cylinder, the file size is just 1/20 of the VRML file produced in the commercial package.

A number of software packages specifically developed for VRML such as 'CosmoWorlds' or 'Spazz' for Avatar development offer good tools for lighting, viewpoint definition and the creation of animated behaviour. For complex geometries, however, special 3D modelling software is needed.

colorSpace.wrl: 305KB

In special cases it can be more efficient to write a java program that generates the VRML file. The example on the left is the result of such a program. It generates colour spaces that update the background colour depending on the position within an RGB cube. Colour emitting lines help with the orientation in the cube. In the java program it is possible to specify the number of subdivisions - here the number specified was 10.


Concrete Examples

Almost all of the VRML examples below have been authored in a simple text editor. They are optimised for the 'CosmoPlayer' running on SGI hardware, but have also been tested in the 'CosmoPlayer' version for the Window's NT platform.


Static Architectural Examples

cafe.wrl: 160KB
& textures

During my IASPIS residency prior to my guest research position at the Interactive Institute I computer simulated parts of the Art and Communication building 'K3' of the Malmo University for a particle systems animation. In order to convert it efficiently into a VRML file each piece of geometry was exported individually and the whole scenario re-assembled by cutting and pasting it in a word editor. The model is static and contains no animated behaviour.
Detail of the K3 cafe area as a meeting place in a blaxxun virtual community.

cafePlus.wrl: 210KB
& textures

The same model has been used as a site for an architectural insertion consisting of wall fragments from 2 thin-walled cubes and distorted spheres copied to the location of particles contained within the thin-walled cubes. In addition to the angled cube fragments a number of distorted spheres have been added. Whenever a particle, which had been emitted radially from a location near the fire extinguisher, travels through one of the cube walls, it becomes the center of a distorted sphere. The spheres can thus be interpreted as visualisations of a radially expanding, yet invisible system. Whilst it is hard, even in VRML, to evaluate such a complex model visually, alternatives such as rendered fly-throughs or still images simply don't allow for spontaneous urges such as 'let's have a look what it looks like over there ...' and thus never give as good a sense of immersion as does the VRML model.


Simple Process Visualisation

humanComp.wrl: 10KB
& textures

The 'To Be Located' project within the Space and Virtuality Studio had designed and built a 'human computer' offering free soft drinks in exchange for information from teenagers. The basic human - 'human computer' interaction consisted of:
saying 'hello' to the 'human computer',
receiving a map of Malmo,
filling in information,
returning the map to the 'human computer'
and, after satisfactory examination of the information supplied,
a soft drink appears in the tray.
The VRML model simulates the submission of the map and subsequent supply of the soft drink can. Basic components have been modelled in a 3D modelling package and exported individually. The outside was photographed and photoshopped to get correct aspect ratios for texture mapping. The entire model, modelling, texturing and animation was completed in approximately 4 hours.


Construction Aides

The 'cardboard Mania' project was a 2 week workshop within the 'Material and Virtual Design' program at the Malmo University, School of Art and Communication. The basic task consisted of converting a torso modelled with Nurbs surfaces into a polygonal model suitable for construction from card or corrugated cardboard. One of the biggest problems in this process is the conversion from Nurbs surfaces to polygons - it requires good design skills to find a ballance between representing the original shape and reducing the number of polygons to a level where the manual assembly process can be completed within a reasonable time frame. On the left we have a plain VRML file - the result of a successful conversion.

labelledTorso.wrl: 67KB

Custom software assigns unique point numbers for each vertex point and outputs a labelled VRML file.

unfolded.wrl: 345KB

The same custom software then unfolds the 3D model and produces a labelled VRML file of the 2D cutting and folding patterns. Simultaneously a DXF file is produced for computer controlled cutting and creasing handled by other applications. The numbered VRML files are essential for the assembly of complex models. For large architectural installations it is a great help to have the VRML files available on-site to determine whether a certain crease between two polygons needs to be concave or convex.

reinforcedTorso.wrl: 42KB

By utilising the transparency options for VRML it is possible to visualise internal reinforcements. This particular example shows the reinforcement of one particular torso designed to be 4 meters tall. Again the possibility of having the labelled files available at the site on a labtop is a big advantage.


HAKI Scaffolding System

dramaUnit.wrl: 14KB

For the 'Low Tech' student workshop and subsequent 'To Be Located' exhibition at Malmo University, HAKI sponsored scaffolding equipment which provided the basic structure for 5 separate exhibition units. Having the different units simulated in VRML allowed the students to check which element goes where and how to assemble it. A special 'mouseOver' function returns the technical name of each element simply by moving the mouse pointer over the geometry in question.

hakiLOD.wrl: 14KB

As the scaffolds get larger and more complex the amount of geometry increases significantly. Utilising VRML's 'level of detail' (LOD) node sensibly ensures that even large models still perform efficiently. 'Level of detail' means that a particular element is represented through different geometries depending on the distance to the current viewing position. A vertical post thus has all the details when the viewer is closest, replaces the 3D modelled hooks for attaching the horizontal elements with 2D look-alikes at mid distance, and is reduced to one simple cylinder primitive further away. The amount of polygons that need to be drawn for a model such as the 'dramaUnit' on the left can thus be reduced by a factor 10 and higher.

assembly.wrl: 14KB

Although the Haki system is particularly straight forward to assemble, it has been used here to demonstrate how to construct a basic module. A series of 4 linked files take the viewer through 4 characteristic steps. Animation of the viewpoint and individual scaffold elements show the order in which to assemble and how to position the elements. The 'mouseOver' function described above is available at all times. A 'EXAMINE' button allows to change the navigation mode and explore the model from angles not covered by the viewpoint animation. The 'NEXT' button loads in the next step starting with the exact position in which the previous one ended. Close-ups, such as STEP 3, can highlight important safety issues such as securing the safety catches. Again, all the models and the VRML browser software can easily be run on an on-site labtop computer.


Visualisation of complex interaction patterns

interface.wrl: 30KB

Within the 'To Be Located' exhibition the 'videoUnit' incorporated a complex interface for accessing video segments from interviews with 4 teenagers. Each video segment represents a teenager and a particular location s/he frequently visits within Malmo. The teenagers are represented through differently coloured puppets; Malmo is represented through a stylised map showing 12 themed squares such as 'home', 'church' or 'cafes'.
By lifting one of the puppets the AV projector overhead highlights potential places favoured by the respective teenager. Placing the teenager at one of the highlighted areas triggers the respective video segment which is then displayed in the top area of the interactive table. Placing the teenager in a non-highlighted area has no effect. Repeating the procedure with another puppet triggers a new video segment replacing the previous one.
The VRML file on the left is a visualisation of this interface prototype. For the web version some details have been simplified and the number of video segments has been reduced.




VRML browsers are generally available as plug-ins into standard web-browsers and are available for downloading at no cost.
Simple VRML authoring software is also available relatively cheaply.
Whilst a graphic accelerator card makes VRML more enjoyable, efficiently authored models perform well on todays standard machines.
VRML performs well through the world wide web. Download times are much faster than sound or movies.
Complex VRML models can be authored in a simple text editing program.
All VRML documentation needed is available on the net. Similar to HTML, VRML can be learned through analysing downloaded source/models and gradually replacing code sections with one's own code/models.
At the high-end of VRML applications, it is possible to utilise, for example, Silicon Graphics standard stereo render mode for active liquid crystal VR glasses, which, if the model is projected onto large-scale screens, rivals other high-end VR technologies.



The selection of good VRML authoring software is very limited.
Unless specialised, as for example software aimed at avatar design in virtual worlds, most VRML authoring software does not produce optimised files.
Export tools in 3D modelling software is generally inadequate. As one frustrated architect commented:

"... you find the possibility in different programs today, such as ArchiCad and Lightscape, to export VRML models. But that is only theoretical. When trying to export an office (without too much detailing) of 300 square meters, the file size got 17,5 Mb. Besides that, it didn't look very good either. You might say then, that I didn't try to optimize the model, for example by replacing mesh elements with textures, but if the work is to be efficient, there is no time for such measures in the daily work." (Janne Jacobsson - email interview 11/99)

Lighting in VRML models is generally quite poor unless the model has been pre-rendered and the geometry fitted with the pre-rendered textures. VRML does not allow for reflections or shadows.



Theoretically VRML knows no limits. In practice the authoring of non-trivial VRML files requires a unique set of skills: Mainly 3D modelling concepts, basic computer animation experience, a good grasp of World Wide Web requirements and a preparedness to familiarise yourself with the VRML specifications. With this set of skills it is possible to generate models efficiently in order to communicate certain spatial arrangements and visualise interactions and processes in simulated 3D computer space.
Similar to HTML, it is possible to aquire the basic skills of creating a simple model yourself; for high-quality products, however, specialist companies can produce more professional solutions. With standard personal computers now being able to run VRML and with increasing knowledge of the possibilities of VRML, it is probably the right time to look seriously into the technology - especially for researchers dealing with spatial arrangements and interaction models.


Horst Kiechle - Malmo, Sweden - November 1999