Real-Time Physics Simulation Forum

Static planar heightfields are fairly popular in collision detection.
However you can extend this idea for convex-convex collision detection/penetration depth using cylindrical or spherical heightfields.

Kind of the 2D version of signed distance maps. Signed distance maps require 2 representations, a 3D map and features (vertices,edges etc).

One vaguely related link I could find is this one: http://gamma.cs.unc.edu/HTextures/Otadu ... nVIS04.pdf

Does anyone have experience, papers or anything in this area?

I don't have any real experience or links to paper on this, but as far as anything goes, a simple thing which could possibly be semi-useful could be to have a cube heightmapped, perhaps projected to fit a sphere or something like that. It should seemingly support convex (semi-concave?) objects without too much trouble?

Erwin Coumans wrote: However you can extend this idea for convex-convex collision detection/penetration depth using cylindrical or spherical heightfields.

Do you mean using cylindrical/spherical coordinates instead of cartesian? How would this help?

Otherwise:
A heightfield maps distances from a 2D plane (perhaps usually a fixed floor, but could be arbitrary). Spherical and cylindrical heightfields map distances from a point or line-segment, respectively. Support for multiple types seems useful, but I would personally hesitate to call them extended cases of heightfields from planes. The heightfield extensions I have seen before do include signed distance fields from each triangle of a tri-mesh. Also signed distance fields from point-sets are used to define some point-set surfaces.