Continuing upward through the questions above, triangles are used because they are the simplest for algorithms to handle, especially if implemented in hardware. GPU implementations tend to be biased toward smaller representations so that a compact list of instructions can be applied to numerous model parts in parallel. It is certainly possible to use more complicated primitives, such as quadrilaterals, splines, and semi-algebraic surfaces [89,124,225]. This could lead to smaller model sizes, but often comes at the expense of greater computational cost for handling each primitive. For example, it is much harder to determine whether two spline surfaces are colliding, in comparison to two 3D triangles.

Steven M LaValle 2020-11-11