Overview / Usage

One of the aspects of the research going into the project was to attempt to construct the underlying level terrain geometry as a boundary represented mesh (as opposed to the more generally used heightfield approach). This because it allows the artist to create more interesting and varied landscape features, such as caves, overhangs, and other more vertical oriented structures.

This has an impact on how rendering, collision detection and pathfinding are handled within the game/simulation. The goal then was to let each of these systems derive and fill out their particular data structures from a singular input mesh (instead of constructing separate nav meshes etc.).

Methodology / Approach

Current approach:

The terrain cage mesh is constructed in Modo from roughly 10k quadrilateral polygons/patches. Vertex meta-data consists of a single UV map and several weight maps used for texture splatting.

For rendering, the mesh is first smoothed using Catmull Clark subdivision and drawn using an adaptive tessellation scheme. Terrain detail is added via a combination of displacement and normal maps.

Collision and pathfinding use the mesh re-meshed at a lower detail level - by only adding detail in high displacement regions, the mesh used for physics matches up with the visualized mesh fairly well even at only 50k tris. As the geometry still forms a closed manifold mesh after this optimization, it can be used for fast signed distance/nearest-point lookups.

Technologies Used

• Custom DX12/C++ based engine
• Uses the GROW computation geometry library for lightmap baking, remeshing, foliage generation etc.

Other links

• Project gallery/journal