The majority of geometric primitives implemented in BRL-CAD are implicit primitives - that is, they are descriptions of volumes and not surfaces enclosing volumes. (A concrete example would be storing a sphere as a center and a radius, rather than a collection of triangles that describe the surface of the sphere). This representation is extremely compact and good for guaranteeing solidity, but has distinct disadvantages when it comes to interactive shaded displays and conversion to other formats.
There has been a lot of work done in BRL-CAD to support representing implicit primitives as NURBS boundary representations. This increases the range of options available for conversion, and sets up possibilities for other features (like 3D shaded displays). We currently have basic implementations for most of our primitives as NURBS, but only located at the mathematical origin and not fully robust (for example, the pipe primitive only works with certain parameter settings, and should work for all pipes).
This GSoC task would be to take the current CSG implicit primitive to BREP conversion routines, complete any missing features (like the ability to represent shapes at locations other than the origin) and make the results robust (valid CSG primitives would generate valid BREPs). The test will be take our example CSG models, convert the CSG primitives to BREPs while preserving the same tree structure, and test to confirm that the raytrace does not change. If it does not, that means the conversion logic has faithfully represented the original data as NURBS data.