In 3D graphics, one of the biggest challenges is creating high quality lighting. Calculating physically accurate lighting is too computationally expensive to be done in real time, so instead multiple approximate models are used to create believable lighting. 

One of these models is ambient lighting, which models indirect light in a scene. The most naïve ambient light model is to have ambient lighting constant across an entire scene. However, this can be greatly improved by ambient occlusion, in which there is less ambient light in areas with occluding geometry – for example in corners, where objects meet, nooks and crannies etc. This vastly improves believability of a scene and makes a scene easier to visually parse.

Computing ambient occlusion can be done in multiple ways. In static scenes ambient occlusion may be computed ahead of time. When this is done, the methods used are typically very expensive and take a long time to compute. However, in dynamic scenes in which content is not predictable ahead of time this is not an option. In this case the state of the art is to use screen space ambient occlusion, which computes the ambient occlusion each frame based on geometry rendered. When this approach is taken an implementation must be chosen which is a good balance of speed and quality.

The CACAO implementation is a highly optimized implementation of ambient occlusion. It may be run at multiple different quality settings, allowing it to meet multiple different requirements for quality performance trade-offs, and be run across a wide range of hardware.