Rather than using a second attractor what I did is add the boundary faces of the non smoothed out mesh to the smooth out mesh. This could be done more than once to add additional rings of faces to smooth out the transition.

I noticed the non-manifold problem, same thing happens in my definition.

Daniel Piker last blog post reminded me of the reason of the non-manifolds.

It's because of the use of cubes. The space has to be packed in a way where coincident edges are shared by only 3 solids (like a voronoi diagram).

Hi Iker -- Have you had any luck in resolving the script? If at all possible would you be able to share your script as I am very interested in how you worked it out and how I could apply it to my agent algorithm. Thank you!!