This week I kicked off an internship with Astrea Media. We began with a general meeting via zoom, with each intern and staff member introducing themselves. I’ll be generating some 3D content for an exciting new project and so far the work is going well. This is not my first experience working with a realtime engine. I’ve done some work in Unity, but never to a polished degree. Our team is working with similar technologies, and I’m eager to learn more about this process.
This week I began optimizations for an existing 3D model. I discovered several hindrances to efficient rendering: absurdly high poly-count, textures in needlessly large image format, non-solid mesh geometry, etc. Rather attempting to rework this model, I instead decided to explore a new design. Due to the organic shapes for this asset, I decided to use Fusion 360’s Sculpt workspace. The sculpt workspace enables designers to create complex shapes with smooth surfaces (e.g., car bodies). There are a few trade-offs to this approach:
Non-parametric design
Sculpt bodies are based on t-splines (I’ll say more about this later), and do not offer design history/timeline functions
Symmetry functions
Rapid mesh generation
Efficient generation organic shapes
T-splines
3D modeling is achieved through a variety of mathematical models for defining shapes. T-splines allow for the creation of freeform surfaces that are defined by a matrix of control points. While meshes tend to be defined by triangles, t-splines work best when all faces are defined by 4 vertices (T-shapes and rectangles).
This last point is important.
This sphere seems like a perfectly logical shape. Indeed, if you click “FINISH FORM” in the upper right corner, Fusion will compute this body and create a solid. Seems simple, right? There’s just one problem: the top of that sphere contains faces with only 3 vertices. As mentioned earlier, t-splines work best when faces are made of 4 vertices. The sphere computes just fine, but as soon as you begin to manipulate this shape, there’s a very good chance that all of those converging 3-point faces at the top and bottom will begin overlapping each other. For example, see what happens when I attempt to apply symmetry to this shape:
Instead of maintaining the converging vertices, the solver calculated something like the iris on a camera.
To avoid problems like this, there’s another option: quadballs.
As you can see, this sphere doesn’t have the same aggressive converging vertices as the other model. The advantage here is that each face can also be split diagonally, efficiently creating triangle mesh faces with minimal distortion.
When exporting this geometry for use in a realtime engine, the mesh conversion produces a high-fidelity representation of the t-spline body, preserving shape details.