Interactive 3D Worlds via the Internet

The problem of presenting interactive 3D environments to a user over the internet is for the most part a tradeoff between bandwidth and processing power. The goal is to create a highly realistic rendition of the 3D environment at the same time giving the viewer the ability to interact with the environment. Interaction can be at many levels from just being able to move around and view the environment, being able to modify and add to the geometry, being able to change material and lighting, being able to interact with other users within the 3D space.

There are two prevalent approaches being pursued on the internet today each with an emphasis on either realism or interaction. The first is the VRML approach where a description of the environment is transfered to viewing software on the users personal computer. After this transfer is complete the interaction of the viewer within the 3D world and the rendering of the views is done locally.

The other approach is to render the current environment remotely and transmit the resulting image to the user. Subsequent interaction (generally movement) is sent to the remote rendering machine/server for processing.

The advantages and disadvantages of these two approaches as well as some other methods will be discussed below.

Local "VRML" model

• The entire geometric description (VRML) is transfered once to the users computer, this results in the most interactive environment.
  
• The viewing software can be designed specifically for the users hardware and operating system interface. It can make optimal use of hardware and fit seamlessly into the operating system interface.
• Modifying the geometry is just as easy as moving around the scene.

• The main restriction is the limit on realism possible given the desire for an interactive experience, for example, each frame needs to be rendered in at least 0.25 seconds. Usually the models are facet based and flat shaded, subtle texture and lighting effects are not feasible.
• Each users computer is different, there are different makes and models, different operating systems and graphics capabilities. This means that it is often difficult to ensure a minimal performance level. It is likely that users will have different experiences depending on their computers capabilities.
• The development of software on multiple platforms can be time consuming. Without multiple platform and operating system support there is the risk of a percentage of the potential userbase not being able to view the material.
• Interacting with other users in the scene is more difficult especially if local users can make geometric modification. All any user movement or interaction needs to be transmitted to all other users who are within the same space, bandwidth constraints limit update times.

Remote server model

• It is quite likely that a specialised rendering server will always out perform user platforms. The possibility of much more realistic renditions can result from both the improved performance but also from software and solutions only feasible or available on higher end platforms. Subtle lighting, texture, and additional geometric complexity greatly enhance the virtual 3D experience.
• It is more likely that the experience users have will be the same irrespective of the platform they are using. "Standard" graphical front ends such as WWW browsers are available for the vast majority of users and present in most cases the same quality on different platforms.
• Interacting with other users in the 3D space should be easier as the the one server "knows" what all the users are doing.

Disadvantages

• The main drawback to remote rendering is bandwidth. Few users today can receive a reasonable quality image from a server in say 0.25 seconds.
• There are different types of interaction, while moving around the environment is straightforward, interactively modifying the scene or changing the lighting is problematic.

Other solutions

This involves rendering all the views the user is ever going to see and storing them as images. CGI programs or static HTML files manage the users movement though the space presenting the appropriate image at the right time.
The primary advantages are: the views can be rendered with the most realistic techniques available and the geometry/textures/lighting can be precisely controlled; there aren't any CPU requirements for the server; the designer is assured that everyone experiences the same quality view.
Of course it isn't possible for the user to change the environment, also the degree of movement is discrete (the finer the steps the more storage space is required for the view database)
An example of this can be seen in the Brain Dynamics Gallery, the fractal gallery, and a description of an earlier experiment called uview.

There are many techniques which would allow higher realism than standard VRML and have lower bandwidth requirements than totally server based rendering systems. One such system is based around a method of being able to interactively view in any direction from a given point in space by precomputing 6 images. The server then computes 6 images each time the user changes position, the 6 images are sent to the user where they can smoothly look around.
This can present high quality renderings, the user can view in any direction in real time. On the down side there is still discrete movement and download delays when the view position is changed and it isn't readily possible to allow the user to interact with the geometry making up the environment.
For more details on the particular technique discussed above see CubeRender. This approach is similar to a system based around Apple QuickTime VR technology where there are also discrete view positions and continuous view direction. In general QuickTime VR environments are precomputed.