SimpleMotionRetargeting

A simple and clean motion retargeting repo extracted from the H2O motion retargeting module

Requirements

1. You need to download the SMPL model from SMPL website and place it in the assets/body_models/smplx directory. The structure should look like this:

   assets/body_models/smplx
   └── SMPLX_NEUTRAL.npz
   └── SMPLX_MALE.npz
   └── SMPLX_FEMALE.npz
   

2. Download the AMASS dataset from AMASS website.

3. Install the required packages:

   pip install -r requirements.txt

   Install pytorch3d package:

   pip install --no-build-isolation git+https://github.com/facebookresearch/pytorch3d.git

Usage

1. Create a config file in the config directory. You can refer to config/orca.yaml for an example.

   • because the SMPL model is in T-pose, and the root coordinate is y-up/z-forward by default, you need to adjust the 'smpl.rotation' to match your robot root coordinate system.
   • you should specify the joint angles in the 'mujoco.T_pose_joints' field in the config file according to your robot model.

2. Fit the SMPL shape to the robot by running:

       python 1_smpl_shape_fit.py config/your_config.yaml

3. Retarget motion by running:

       python 2_smpl_motion_fit.py config/your_config.yaml --data path/to/your/amass/motion.npz

4. Visualize the retargeted motion by running:

       python vis_mujoco.py config/your_config.yaml --data retargeted_data/path/to/your/amass/motion.pkl