I'd like to describe two common issues that we encounter with collisions between adjacent sub-meshes, and propose two related feature that will address one or both of them. I don't think either of the issues is in any way MuJoCo-specific.
Issues:
Penetration of thin objects between two adjacent sub-meshes: When a thin object penetrates between two adjacent meshes, the contact forces push sideways rather than out, leading to jamming of the thin object between the sub-meshes.
Catching on corners: When a pointy (not necessarily thin) object slides along the side of a decomposed mesh, the corner can catch on the corner of one of the decomposed meshes.
Proposed solutions:
"Inflate" all sub-meshes in the direction normal to internal mesh-adjacent surfaces by an overlap parameter $d$, with units of length.
Make the added (inflated) part smooth, by adding several faces that attempt to approximate a smooth bevel at the corners. This requires specifying the smoothing radius $r$ and possibly also the number of added vertices/faces that participate in the smooth corner.
I'm attaching an image below with a sketch of what we are imagining. Note that for us issue 1 is much more significant than issue 2, and the proposed solution 3 is both easier to implement and does not introduce new faces/vertices. However note that solution 4 is more general, since it solves both issues 1 and 2 (but cannot be said to be strictly better, due to the increase in the number of faces).
We'd love to hear your opinion on these proposed features.
Hi CoACD,
I'd like to describe two common issues that we encounter with collisions between adjacent sub-meshes, and propose two related feature that will address one or both of them. I don't think either of the issues is in any way MuJoCo-specific.
Issues:
Proposed solutions:
I'm attaching an image below with a sketch of what we are imagining. Note that for us issue 1 is much more significant than issue 2, and the proposed solution 3 is both easier to implement and does not introduce new faces/vertices. However note that solution 4 is more general, since it solves both issues 1 and 2 (but cannot be said to be strictly better, due to the increase in the number of faces).
We'd love to hear your opinion on these proposed features.
cc @quagla, @btaba, @erez-tom.