Marching Cubes Cave Generation

• 8 Weeks, 50%
• Unreal Engine

I have always been fond of procedurally created worlds and the pure sense of adventure they spark. They can be the defining factor which makes a repetitive game interesting; breaking up the monotony of playing the same level and tile set time and time again.

While there are many different techniques in the world of PCG, I chose Marching Cubes. I deemed it fitting for the given time frame, and challenging enough so I'd actually learn something from it. This also led me to choose Unreal as the development environment.

I had minimal prior experience in Unreal and saw it as an opportunity to gain knowledge in one of the premier Development Environments on the market.

Procedural Generation: while hand crafting unique environments may fit for most gameplay experiences, some simply need a bit of randomization.

Using Procedural Generation we can give the Players a more varied gameplay loop with a new environment every session or adventure. This technique is often used in Session-Based-Games or Open-World-Games to provide a low-poly look but can be customized further to give a more realistic look. It can also be used by developers as a base for creating a level, which they then customize and decorate as they see fit.

The Algorithm "marches" through the volume with a cube and samples values from the volume in its 8 corners, and then determines what mesh should be created there to connect all the dots.

It does this by referencing several look-up-tables which contain all the necessary data needed to create a mesh inside the current cube.

This process continues until the entire volume has now been filled with a Mesh.

A simple 0-1 value determining how smooth the generated Mesh is.

By inheriting from the same base class we can easily create and switch what complexity of Noise we're using. The Project includes a Simple 3D Perlin Noise implementation and a Blended version, where a Noise is used to blend between 2 different Perlin Noises.

Also inheriting from a base class, the User can create different objects with functionality to be created on chosen surfaces by specifying an interval between two rotations. The Project features Stalactites and Crystals.

A Shader implementation in Unreal's Shader Graph which lets the User choose what each cave surface should look like.

Originally, I did have a fun but way more complex idea for a Cave Generation Tool. It was ultimately turned down for much the same reasons as the Event Graph implementations; time constraint.

It would have been built around the user (or an algorithm) generating curves in the world, and splitting them into intersections wherever they wished. They would then be able to press a button to generate a Cave Mesh around this curve using the Marching Cubes algorithm.

This version lends itself to more mission based Games, where we can place desired rooms in the world and then connect them with the serpentine tunnels generated by the Tool. As well as being a more Tool-like implementation which is way more friendly towards non-programmers than the current system.

After setting up and planning the Project in Week 1, I could finally start with working on the algorithm itself.

By retrieving a 3D perlin noise value from FMath::PerlinNoise3D, and deciding upon a floor value, we can determine what parts of the cube grid should be air and ground. As seen in the following image.

While the prototype method of using lots and lots of cube meshes worked until now, the performance was getting a bit out of control.

Restructuring
Having researched the subject extensively I began working on my own implementation. The mesh logic was optimized by structuring everything into a single Procedural Mesh Component per Chunk. This meant that some much more complex math for Vertices and Indices had to be done.

Luckily most of these problems are solved by publicly available Look-Up-Tables which most Marching Cubes implementations use. These tables tell us how to generate our vertices and their edges.

By continuing this and tinkering with other parameters and implementations such as smoothing, I eventually landed on this result:

The smoothing logic works by manipulating the edges of each "cube" in the Marching Cubes, moving them based on the delta of their connections (from a lookup-table) and the cutoff value from the Noise itself. It then multiplies the value by the exposed Smoothness variable, allowing the user to linearly choose how jagged or smooth the surfaces become.

Noise Scale
This was pretty convincing already, having completed the general look of the Marching Cubes Algorithm already.

But it's still far off from being an actual cave. By manipulating the size of the noise on the Y-Scale we can squish it together a bit, leading to a more cave-like structure.

Now that the basic shape of the caves was out of the way, I set my sights on making it look more cave-like.

This started out with something called "Triplanar-Shaders", which are shaders which can map the given textures based on the world axis' instead of the objects normals. They work much like a cubemap where you can sample a texture from a 3-dimensional direction.

What this means in practice, is that we can differentiate which texture should be displayed on each world axis, and therefore have different textures depending on if it's the roof, walls, or floor. Using such an effect we can effectively give life to our procedural flat surface.

Achieving this result:

(Click to Enhance)

This was not the end of the Triplanar Shaders, as I made multiple returns to improve them over the Project's course.

If you paid attention you'll notice that the Caves suddenly have Stalactites! While the object generation system was started during this week it will instead be covered in the following chapter, where it was completed.

This week was mostly spent on Polishing and improving the UX & UI, while also implementing extras such as Infinite Generation and Mesh Deformation.

These weeks were mostly spent on this Website or my Resumé, and will therefore not be covered. These were however the Weeks that Mesh Deformation was completed.