how to generate a new t-pose file from a customized humanoid

[RavenLeeANU]

Hi team,

I want to add a new humanoid for retargeting, but generating a new t-pose file from the urdf file seems not included in the pipeline. so could you add some hints for preprocess in the tutorials? thank you very much.

[ZiangZheng]

Hi, thanks for your question!

Currently, we recommend using open-source retargeting tools such as GMR to generate the retargeted data — this approach is generally more robust and flexible.

If you prefer to use the poselib-based retargeting pipeline, you can refer to the following scripts:

trackerLab/scripts/poselib/tpose/tpose_mjcf.py — for generating the basic zero pose from your URDF/MJCF model.

trackerLab/scripts/poselib/tpose/tpose_mod.py — for converting the zero pose to a standard T-pose and removing unmatched joints.

We'll also consider adding a short preprocessing tutorial for new humanoid models in future updates.

Hope this helps!

[geoffreygao1]

Hi,

Still a beginner to this workflow, but one of the big benefits of TrackerLab over GMR is that it outputs ready to use .npz motion files for Isaac Lab whereas GMR doesn't have this option. Do you have any advice about converting the retargeted GMR ouput into .npz?

[ZiangZheng]

Hi,

That's a great observation! You are correct; one of the key advantages of TrackerLab is its ability to output Isaac Lab-ready .npz motion files.

We provide a specialized script that can easily convert your retargeted GMR .pkl output into the .npz format suitable for Isaac Lab's tracking utilities. This conversion process performs forward kinematics (FK) to ensure the motion is correctly represented in the simulation.

🛠️ Conversion Script Location
You can find the conversion utility here:

/home/ubuntu/code/renforce/trackerLab/scripts/data_tools/data_fk

🔑 Important Requirements
For this script to work correctly, you need the latest version of TrackerLab, which includes updates to key modules that handle joint alignment and trajectory smoothing:

Joint ID Alignment and Interpolation: Ensure you have the latest versions of:

trackerLab/source/trackerLab/trackerLab/motion_buffer/motion_lib/motion_lib_dof_pos.py

trackerLab/source/deploylib/deploylib/deploy_manager/deploy_manager.py

These files are crucial for correctly aligning the joint IDs and applying appropriate SLERP (Spherical Linear Interpolation) and linear interpolation methods to the motion trajectory data.

Let me know if you run into any issues running the script!