New Reality Engine (XNA)

Date
Spring 2009
Tech
XNA, C#, HLSL, Xbox 360

Fresh Start

Naturally, version one of anything becomes a lesson in how not to do things, and it's not long until you look upon your codebase with regret as to the monster it has become.

Despite really enjoying OpenGL, I could not help but be curious as to how things were on Microsoft's side of the fence. I wasn't too keen on C++ with DirectX, but XNA just seemed to be a fantastic balance of "here's how you should be doing it" and "the ball is in your court now".

Looking at the screenshots below, you'll instantly notice a drastic improvement of the visuals compared the OpenGL version. The same techinques are being used, but the OpenGL version did not have texture mipmapping enabled, which despite being a detail culling technique, actually causes textures to look grainy and false.

This particular project went as far as terrain, shaders, L-Systems, and Kevin Boulanger's procedural grass technique.

Terrain ported over. No lighting, and the horrible looking texturing thanks to no mip-mapping.
Terrain ported over. No lighting, and the horrible looking texturing thanks to no mip-mapping.
Per-vertex lighting.
Per-vertex lighting.
Adding a skydome, currently of which is not following the camera.
Adding a skydome, currently of which is not following the camera.
Skydome, and lighting with low values.
Skydome, and lighting with low values.
More of the same..
More of the same..
A few tweaks to the shader, and brighter values.
A few tweaks to the shader, and brighter values.
Added specular lighting. The tiny 'hairs' are the vertex normals.
Added specular lighting. The tiny "hairs" are the vertex normals.
A comparison of the heightmap, and generated normal map.
A comparison of the heightmap, and generated normal map.
An accident, but this looked like water caustics, albeit static.
An accident, but this looked like water caustics, albeit static.
A more sensible lighting setup.
A more sensible lighting setup.
With multitexturing ported in. Looks a bit silly.
With multitexturing ported in. Looks a bit silly.
More of the same..
More of the same..
Passes..
Passes..

L-Systems

With terrain being procedurally generated, one idea was to take it to the extreme and have the textures, and as many models as possible procedurally generated too. The end-goal would be a game engine that could fill a few MB at most, but display fairly convincing environments.

No, I didn't reach that goal.

L-systems is an algorithm that allows us to generate predictable patterns, such as those you'd find on a typical tree. We draw a tree trunk, then at the endpoint of the trunk, we draw a branch. We then roll back to the first endpoint, and draw another branch at a different angle. We then repeat these steps with some random offsets, and recurse the algorithm, and we soon have the outline of a tree.

That's the theory anyway. Add some rotation offset to make it 3D, play around with the values enough, and the results should be passable.

Variant 1 - each recursion generating smaller steps of branches.
Variant 1 - each recursion generating smaller steps of branches.
Variant 2, with more loops of each step.
Variant 2, with more loops of each step.

Back to projects