Question About Specific Narrowphase Collision Algorithm

[Cfather]

Question About Specific Narrowphase Collision Algorithm

[lmontaut]

Hi @Cfather,
For box-sphere, both hpp::fcl::collide and hpp::fcl::distance will call the specialized function in src/narrowphase/details.h.
This specialized function is faster than running GJK+EPA on box-sphere, notably due to the fact that a sphere is strictly convex. This means that in this case GJK and EPA will do more iterations compared to a polytope-polytope collision pair, because these algos will try to "fit" the sphere with polytopes, which is costly.

At the moment, box-sphere takes about 40 nano-seconds with the current specialized function. If you instead use a box-ellipsoid (with radii = (r, r, r), r being the sphere radius) collision pair, which will use GJK+EPA behind the scene, you will be in the range of 0.5 to 1 micro-second (depending on whether or not the shapes are in collision).

[Cfather]

Thanks for the comments. This is very clear.