Procedural Content, and Aiming Too High

At the outset of this project my initial aim was to allow full control over the entirety of a scene, as far as level design and editing is concerned. After sitting down and carefully considering potential routes of implementing functionality to allow users to create custom worlds and other assets, I've come to the decision that the work required to provide an interface for editing such assets will only hinder my ability to actually complete this project.

With my previous project, Revolude, it seemed rather slick to procedurally generate the world when the game starts. Whoever was responsible for launching the game server had access to various sliders and coloration options they could play with to customize the 'style' of the world that people would experience while playing in the game they are hosting.

the "start game" menu screen from my previous project 'Revolude'

This functionality alone was simple enough to implement, and was (to my mind) the happy medium between designing your own level, and choosing an existing one from a list when launching a game. A world seed value (not exposed via the UI menu screen) along with all of the server's world-generation parameters are gathered up and sent off to connecting clients who are joining the game, so that they can generate their own local copy of the entire world for rendering and collision detection prediction.

One of the primary advantages to this approach is that every server can be running a completely unique world without clients being required to download or store any map geometry from the server. The only thing transferred are some parameters for the procedural generation of the world. This is, in my mind, the great advantage to using procedural methods.

What with my non-existent experience concerning floating point determinism, I did run into some serious bugs where trees would sometimes be planted, or not, in specific 'close call' spots that seemed too steep to some CPUs, and not too steep on others. This resulted in worlds that were somehow slightly different on two machines that were playing in the same server. These sorts of issues are resolvable, especially if you are aware of them before writing any code in the first place.

Nonetheless, I really like the idea of providing a user with procedural tools to create a base scene volume, from which they could construct their vision by hand. These worlds would be saved, in compressed form, to disk, and would be uploaded to connecting clients. I envisioned a full built-in editor for flying around sculpting worlds, placing entities and detail props, etc.. Along with a procedural materials editor, an entity voxel volume editor, and possibly a synthesized sound editor (Revolude actually featured console-scripted sound synthesis, with enveloping and a few effects, but required hand-editing sound parameters in an external text editor and alt-tabbing back and forth to listen to the resulting sound).

It just seems like too much work for what I'm trying to accomplish. What I would rather do is give a preview to a user that is starting a game. This would just be a more advanced version of what Revolude does. Storing to disk, and transmitting actual compressed volume data to clients just sounds too expensive, and defeats what I initially hoped to accomplish by utilizing procedural methods in the first place.

No comments:

Post a Comment