How It Works
From a high-level, the Volumetric Lightmaps system works in the following manner:
Lightmass places lighting samples throughout the level and computes indirect lighting for them during the lighting build.
When it comes time to render a dynamic object, the Volumetric Lightmap is interpolated to each pixel being shaded, providing precomputed indirect lighting.
If no built lighting is available (meaning the object is new or has moved too much), lighting is interpolated to each pixel from the Volumetric Lightmap for Static objects until lighting is rebuilt.
When a Lightmass Importance Volume is placed, the Volumetric Lightmap builds bricks that are made up of 4x4x4 cells (lighting samples). When lightmass is run, the cells are placed over the whole Lightmass Importance Volume and then it uses more cells around static geometry in the scene to capture better indirect lighting results.
Each of these points (or spheres) is a Volumetric Lightmap lighting sample, which uses a Third Order Spherical Harmonic to store the incoming lighting in all directions.
Near objects, any static geometry that is within one of the bricks will use more cells where indirect lighting changes the most. This data structure enables interpolation of indirect lighting to any point in space on the GPU.
(From left to right) A Lightmass Importance Volume with a single Static Mesh placed within the volume. Cells around the static geometry are higher-density; An example representation of a single side of a brick showing the placement of 4x4x4 cells; The same example representation of cells with higher-density when static geometry is present during a lighting build.