I don't think that establishing a point based grid is very efficient.
What should I do in order to achieve very precise pathfinding, i mean to a level that every meter has to be catered for...
Thanks
Jack

I use recast for space partitioning, but there are a lot glitches all over the place.

For most/general purposes the approaches mentioned above are fine. A tight grid or navmesh can do wonders. If you're keen and have a few spare months you could do your own navmesh implementation using BSP to avoid voxelisation errors. Discussing your project/problem may be a good idea, there may be workarounds that give you more bang for your buck than an extremely finely grained pathfinding system.

I've heard a lot of good things about recast. I would suggest investigating something like the below. Recast is a complex beast, and tweaking the settings may give you better results.
http://digestingduck.blogspot.co.nz/2009/08/recast-settings-uncovered.html

I have personally developed a C++ program with max SDK, that does this
- cast a ray from every pixel of the ground upwards and if it hits anything, it is a blockade, otherwise not
- this results in a pixel-based grid without having to worry about voxelisation errors at all.

The problem is
The size of one object doesn't match the size of one unit of the map
Thanks