This project requires Isaac Sim and Isaac Lab installed via pip. If you haven't installed them yet, pleanse follow the instructions in Isaac Lab Installation Guide
No symbolic links or special setup required! This project works as a standalone Isaac Lab extension.
Install the repo in editable mode (mirrors the workflow used in unitree_rl_lab):
pip install -e .We copy the train and play script from isaaclab, note you do not need to do it yourself.
bash ./sync_skrl_scripts.shpython -m humanoid_amp.train --task Isaac-G1-AMP-Walk-Direct-v0 --headless --num_envs 4096or for dance training:
python -m humanoid_amp.train --task Isaac-G1-AMP-Dance-Direct-v0 --headless --num_envs 4096Additional training options:
# Resume from checkpoint
python -m humanoid_amp.train --task Isaac-G1-AMP-Walk-Direct-v0 --checkpoint logs/skrl/path/to/checkpointpython -m humanoid_amp.play --task Isaac-G1-AMP-Walk-Direct-v0 --num_envs 32 --checkpoint logs/skrl/<run>/checkpoints/Latest.ckptpython -m tensorboard.main --logdir logs/skrl/Then open your browser to http://localhost:6006
Walk training: master branch. Dance training: dance branch.
The usage of some helper script functions is explained at the beginning of the file.
motion_loader.py- Load motion data from npz files and provide sampling functionalitymotion_viewer.py- 3D visualization player for motion datadata_convert.py- Convert CSV motion data to npz format with interpolation and forward kinematicsmotion_replayer.py- Replay motion data in Isaac Sim with optional recordingrecord_data.py- Recording and managing motion data utility classesverify_motion.py- Verify and display npz file contentsvisualize_motion.py- Generate interactive HTML charts to visualize motion dataupdate_pelvis_data.py- Fix pelvis posture and body center issues in replay motion by replacing pelvis data from source file
Note: The original dataset and URDF files from Unitree Robotics have been removed by the official source.
If you're still looking for the dataset, a third-party mirror is currently available here:
lvhaidong/LAFAN1_Retargeting_Dataset
Or you can also use the data from AMASS
- ✅ Current: Dancing motion is working
- ✅ Test the
testbranch thoroughly and merge it intodanceas soon as possible - ✅ Write a more detailed README to cover the new features and usage
- Add clearer comments and explanations in all Python scripts
Contributions, discussions, and stars are all welcome! ❥(^_-)
