Create T-Pose .npy with Poselib

Hi,

I’m failing to create a T-Pose .npy. I want to retarget the SFU Wushu motion to the AMP skeleton. As I do not have any MJCF file from them but only FBX, I tried to extract the T-Pose from the FBX and saved it as another FBX. However, when I run fbx_importer on it it creates a SkeletonMotion but the retargeting expects a SkeletonState for the T-Pose. In the example a SkeletonState is only created from a MJCF file however.

Best,
Noshaba

Hi,
Any updates? I am also stuck on generating t pose using non-CMU sources

I just asked the Isaac Gym folks to provide me the T pose file ^^’

Does anyone have an answer to this by now?

You can create a function based on SkeletonMotion.from_fbx, you will notice it generates an SkeletonTree when not given as a param:

skeleton_tree = SkeletonTree(joint_names, joint_parents, local_translation)

This SkeletonTree can be used to generate a pose. If the fbx file is already in t-pose, you can generate a SkeletonState with a zero rotation pose:

tpose = SkeletonState.zero_pose(skeleton_tree)

Finally, save the tpose in .npy by:

tpose.to_file("data/tpose.npy")