ManiSkill icon indicating copy to clipboard operation
ManiSkill copied to clipboard

More objects of partnet_mobility

Open ZhuFengdaaa opened this issue 1 year ago • 1 comments

In this page, you released 10 thousand partet mobility objects. However, the python tools/download.py --uid faucet command can only download hundreds of them. The data format is different. Faucet task reads the moblity_cvx.urdf file while the released large-scale dataset provide only mobility.urdf.

Could you please release the preprocess script or the processed data of other partnet mobility objects?

ZhuFengdaaa avatar Aug 30 '22 13:08 ZhuFengdaaa

Hi, in order to keep hold-out assets private for the ManiSkill 2022 challenge, we will not release the preprocessing script.

However, I can share the general idea about how to preprocess articulations. The key difference between mobility_cvx.urdf and mobility.urdf is whether the collision meshes are processed to be convex.

  1. Parse URDF (recommended tools: bs4, https://github.com/mmatl/urdfpy, pytransform3d)
  2. For each link, merge all visual meshes, and do convex decomposition (V-HACD, COACD) for the merged mesh to get the collision mesh of the link.
  3. Save the collision mesh and add collision tag in URDF
  4. Save URDF

Jiayuan-Gu avatar Aug 30 '22 18:08 Jiayuan-Gu

The processed assets of ManiSkill 2021 will be released in the next version.

Jiayuan-Gu avatar Sep 03 '22 00:09 Jiayuan-Gu