The problem here is the fact that as far as I know there is no tool (at least available to the public) that supports the full Collada specification (especially 1.5). The part that is in most cases well tested and developed is the geometry-library element, materials etc., which in 1.5 are usually taken from some 1.4 implementation (that's why tools that usually state they support 1.5 actually don't (physics, kinematics etc. are in most cases missing or in bad condition). Still you can easily create decent (as in geometry part is OK but the rest - maybe is, maybe not) enough Collada files using various export-features of primarily 3d modeling software (Blender, Maya, 3ds Max, CATIA etc.). OpenRAVE (used for robotics path planning) actually has one of the best export/import capabilities, when it comes to COLLADA and even supports (partially) 1.5 features such as kinematics.
If you decide to use Blender for example (free and open source so you can actually look how the import/export addon works), you can create something simple or complex and export it as COLLADA 1.4 (not 1.5!). OpenRAVE for instance uses a custom XML-format that is converted internally to Collada (in order to hide the complexity of this standard) and allows you to even embed other formats (mostly for the geometry-part) such as OBJ, which is much much easier to find a decend import/export tool for. Khronos Group actually provides OpenCollada (OpenRAVE and many others use it internally, which of course results in bugs in all of them, when something is badly implemented in OpenCollada :P). The Assimp-library offers also quite a lot but the major problem is the misinformation it gives on what it actually supports from the Collada standard. In fact it is really, really hard to find a reference on the implemented features, when it comes to Collada, and sadly recently I started using 1.5 (kinematics) just to discover that Assimp supports only 1.4 and is bound to it to such extent that it throws errors at you the moment it encounters a typically 1.5 element (even if it is empty!), which omho is a really bad implementation on part of the developers. In the list of supported formats the Assimp's site states only Collada and no version is given.
I know this question is old and answered but I hope this info helps. I myself am writing a parser in C# for internal usage where I'm currently working and it's a real pain to discover how badly supported this already an ISO standard is. The complexity of Collada is huge but that's why it is considered a pipeline-format and not something you are supposed to use in a final product that relays on good performance (both speed and storage).