[Proposal] More configurations for convex decomposition in the mesh converter

[hu-hy17]

Proposal

Allow users to specify more configurations when converting mesh files (.obj, .stl, etc.) to USD files through isaaclab.sim.converters.MeshConverter.

Motivation

When I use MeshConverter with --collision-approximation convexDecomposition option to get a collision model, the collision model is significantly larger than the visual one.

The problem is that the converter uses a very low voxel resolution by default (I think it should be 50000), as shown in the picture (the left one is converted from MeshConverter and the right one is re-decomposed with voxel resolution=5000000).

I tried to solve this by modifying lines 124- 132 in isaaclab.sim.converters.MeshConverter

from pxr import PhysxSchema
if cfg.collision_props is not None:
    # -- Collision approximation to mesh
    mesh_collision_api = UsdPhysics.MeshCollisionAPI.Apply(child_mesh_prim)
    mesh_collision_api.GetApproximationAttr().Set(cfg.collision_approximation)

    # Add the following three lines
    if cfg.collision_approximation == 'convexDecomposition':
        cvx_api = PhysxSchema.PhysxConvexDecompositionCollisionAPI.Apply(child_mesh_prim)
        cvx_api.CreateVoxelResolutionAttr(5000000)  # 50000-5000000

    # -- Collider properties such as offset, scale, etc.
    schemas.define_collision_properties(
        prim_path=child_mesh_prim.GetPath(), cfg=cfg.collision_props, stage=stage
    )

I hope that in addition to voxel resolution, users can also define more options to get a better collision model.

Alternatives

Additional context

Checklist

• [x] I have checked that there is no similar issue in the repo (required)

Acceptance Criteria

Add the criteria for which this task is considered done. If not known at issue creation time, you can add this once the issue is assigned.

• [ ]

[RandomOakForest]

Thank you for posting this proposal. The team will review it.