I'm trying to create a graphics engine (in OpenGL), to render large worlds that are procedurally generated or player influenced. Example, the player may construct a building, say a tower, and from a distance you should be able to see the tower.
When I sat down to think about my scene management, I decided I wanted interaction with the system from the programming point to be pretty simple. I wanted the ability to create a complex object and treat it as a single entity. For example create a torch file where theres the base mesh for the handle, A particle effect for a fire, and a point light to illuminate the surroundings. I wanted adding, manipulating, and removing objects to be accomplished in a few lines. And have all the magic happen behind the scenes.
Example.
GraphicsObject *obj = AssetManager.Get ("SomeObjectFile");World.AddObject (obj);So I immediately thought scene graph, because of how you connect things. However, I don't really like the idea of a scene graph. Most objects in a world don't need so many tiers of control. For example, even a magical sword might only have a Mesh, a light, and a particle effect. They all work well at the same level, no need to create a hierarchy. Also, I always thought a scene node should represent something in the world, but for skeletal animation, you basically need a scene node on each bone(if you want to attach to it), which seems completely unintuitive to me because the bone nodes don't actually render anything they just sit there to have a transform. This bone node concept made it harder for me to picture a system to allow you to specify where weapons and armor attach, you'd probably want to reference a bone by its name like "RHand", the typical scene node doesn't need named children, which means more special cases. It seemed that a scene graph was a messy solution.
This guy reinforced some of my doubt in a scene graph: http://www.gamearchitect.net/Articles/GameObjects2.html
For the sake of the explanation I'll call things like meshes, lights, billboards, and particle effects "Graphics Elements". (They are all derived from a GraphicsElement class in my code anyway).
I thought it would be advantageous to create a "Pipe" structure where it held a linked list of render jobs. The jobs would be held onto by their respective GraphicsElement. So there would be a pipe for Opaque Objects, Translucent Objects, and Lights. The linked list would easily handle insertion and removal, and the pipes would only iterate over the proper parts needed for that section of the scene. Plus GraphicsElements could manage insertion and removal only when they need changing So objects remaining in the scene take no time to reinsert themselves.
So I thought harder about what I needed. Most objects in a fantasy world (my target) are pretty simple, a tree, a rock, a chair, a barrel, are all one mesh no flair objects. But some things need to be batched together for efficiency like trees. So a GraphicsObject should be able to subscribe to different rendering options, either being batched in with similar objects or be thrown straight into the pipeline. They shouldn't waste memory supporting children objects and whatnot. I think they're pretty straight forward to LOD aswell, if they're too small don't draw them, perhaps a progressive mesh for inbetween.
Other objects are more complex, things like torches, fountains, magical spells (projectiles), and enchanted weaponry often have a mesh, maybe a particle effect, a light, and maybe an animation. These things are less common, but still don't really need a tiered system. Most things could just a list and a transform for each GraphicsElement present. Their LOD process is a bit more complicated, obviously various elements shouldn't be used at certain distances, unless they are extremely large, say a bonfire particle effect. Perhaps lower the number of particles and up their individual size. (Probably should be a property of the particle effect, nothing major)
But perhaps, the thing that eludes me most is character objects. Usually at the base would be an animated mesh of sorts, and then weapon objects should be attachable to bones. Bones should support multiple attached objects (I imagined getting pin cushioned by arrows...) My only thought was to create some sort of Bone Slot. A Slot would hold a list of GraphicsElements with transforms, a list of attached GraphicsObjects, and a name. Slots would be sorted by name for easy retrieval(making it easy to specify a sword attaches to the "RHand" bone). So Graphics Elements could be included with a creature with the GraphicsElement list. Example, lantern fish would have a light on his little fishing rod tentacle thing. As well as objects like armor and arrows could be attached to bones. Lod control would just follow down the hierarchy executing their normal LOD method.
So I guess my standing concerns include.
How should a mesh LOD be handled. I knew lower poly versions could be swapped in, but I also read about progressive meshes. I know typically to LOD animation you eliminate deep bones in the hierarchy and lower the interpolations between key frames.
For LOD of an object as a whole I thought about using some sort of radius approximation supplied in the object file. Objects with enormous radius's might use an impostor billboard to show up far off in the distance(like the tower I mentioned).
What do I need to think about for occluding, other than the frustum? I would like indoor environments as well, so maybe portals?
Does this scene management structure look like it would work? Can you see any major problems with it?
My other 3D creations have been quite small and needed no sort of management other than a list that was iterated to render.
Things I looked at:
http://www.insomniacgames.com/tech/articles/0308/files/progressive_mesh.pdf
http://www.gamasutra.com/features/20060105/davis_01.shtml
http://www.gamearchitect.net/Articles/GameObjects2.html
http://www.gamedev.net/reference/articles/article1812.asp
http://www.gamedev.net/community/forums/topic.asp?topic_id=483936
Thanks
~Tocs
