Skip to content

Latest commit

 

History

History
67 lines (50 loc) · 2.39 KB

retargeting.md

File metadata and controls

67 lines (50 loc) · 2.39 KB
Raw

Retargeting to Your Own Humanoids

To retarget SMPL data to your own humanoids, you need to follow the following steps:

First, download and setup the SMPL body models as well as AMASS data (code is provided in the README.md).

Then, run the following scripts to fit the SMPL motion and shape to your humanoids. We will use unitree_g1 as an example.

First, we will find the SMPL shape parameters that best fit the humanoid.

python scripts/data_process/fit_smpl_shape.py robot=unitree_g1_fitting

Then, we will find the SMPL motion parameters that best fit the humanoid. (pass in +fit_all=True to fit all the motion clips in AMASS, otherwise only fit one motion)

python scripts/data_process/fit_smpl_motion.py robot=unitree_g1_fitting +amass_root=/path/to/your/amass/data

To visualize the fitted motion, you can run the following script.

python scripts/vis/vis_q_mj.py robot=unitree_g1_fitting +motion_name=0-Transitions_mocap_mazen_c3d_dance_stand_poses

Details about the YAML files:

extend_config:
  - joint_name: "head_link"
    parent_name: "pelvis"
    pos: [0.0, 0.0, 0.4]
    rot: [1.0, 0.0, 0.0, 0.0]

base_link: "torso_link"
joint_matches:
  - ["pelvis", "Pelvis"]
  - ["left_hip_pitch_link", "L_Hip"]
  - ["left_knee_link", "L_Knee"]
  - ["left_ankle_roll_link", "L_Ankle"]
  - ["right_hip_pitch_link", "R_Hip"]
  - ["right_knee_link", "R_Knee"]
  - ["right_ankle_roll_link", "R_Ankle"]
  - ["left_shoulder_roll_link", "L_Shoulder"]
  - ["left_elbow_pitch_link", "L_Elbow"]
  - ["left_zero_link", "L_Hand"]
  - ["right_shoulder_roll_link", "R_Shoulder"]
  - ["right_elbow_pitch_link", "R_Elbow"]
  - ["right_zero_link", "R_Hand"]
  - ["head_link", "Head"]


smpl_pose_modifier:
  - Pelvis: "[np.pi/2, 0, np.pi/2]"
  - L_Shoulder: "[0, 0, -np.pi/2]"
  - R_Shoulder: "[0, 0, np.pi/2]"
  - L_Elbow: "[0, -np.pi/2, 0]"
  - R_Elbow: "[0, np.pi/2, 0]"

The above YAML file is used to setup the retargeting process.

Extend_config: creating new joints for the humanoid such that there can be more constraints for the motion fitting process (e.g. the head joint is created to be a child of the pelvis joint)

Joint matches: matching the joints between the humanoid and the SMPL model.

SMPL pose modifier: modifying the SMPL pose to fit reseting pose of the humanoid. For instance, H1's default pose has the elbow bent, so we need to modify the SMPL pose to fit the humanoid.