This repository contains the official implementation of Morph. Morph is a Motion-free physics optimization framework, comprising a Motion Generator and a Motion Physics Refinement module, for enhancing physical plausibility without relying on costly real-world motion data.

Download Isaac Gym from the website, then follow the installation instructions.

Once Isaac Gym is installed, install the external dependencies for this repo:

pip install -r requirements.txt

First, an Morph model can be trained to imitate a dataset of motions clips using the following command:

horovodrun -np 8 python ase/run.py --task HumanoidLocationSMPL --cfg_env ase/data/cfg/humanoid_smpl_location.yaml --cfg_train ase/data/cfg/train/rlg/smpl_humanoid_task.yaml --motion_file ase/data/motions/WalkDog.npy --headless

--motion_file can be used to specify a dataset of motion clips that the model should imitate. The task HumanoidLocationSMPL will train a model to imitate a dataset of motion clips. Over the course of training, the latest checkpoint Humanoid.pth will be regularly saved to output/, along with a Tensorboard log. --headless is used to disable visualizations. If you want to view the simulation, simply remove this flag.

To test a trained model, use the following command:

python ase/run.py --test --task HumanoidLocationSMPL --num_envs 1 --cfg_env ase/data/cfg/humanoid_smpl.yaml --cfg_train ase/data/cfg/train/rlg/smpl_humanoid_task.yaml --motion_file ase/data/motions/WalkDog.npy --checkpoint output/Humanoid_19-02-01-38/nn/Humanoid.pth

Motion clips are located in ase/data/motions/. Individual motion clips are stored as .npy files. Motion clips can be visualized with the following command:

python ase/run.py --test --task HumanoidViewMotion --num_envs 2 --cfg_env ase/data/cfg/humanoid_smpl.yaml --cfg_train ase/data/cfg/train/rlg/smpl_humanoid_task.yaml --motion_file ase/data/motions/WalkDog.npy

--motion_file can be used to visualize a single motion clip .npy

If you want to load new motion clips to the character, you can take a look at an example script in ase/poselib/load_motion.py.

If you want to visualize the motion as .fbx, you can take a look at an example script in SMPL-to-FBX/Convert.py.

The provided code and pre-trained weights are licensed under the Apache 2.0 license.

This code is based on ASE. Some code are brought from EDGE. We thank the authors for their awesome work.

If you have any questions, please feel free to reach me out at [email protected].

If you find our work useful for your research, please consider citing our paper:

@article{li2024morph,
  title={Morph: A Motion-free Physics Optimization Framework for Human Motion Generation},
  author={Li, Zhuo and Luo, Mingshuang and Hou, Ruibing and Zhao, Xin and Liu, Hao and Chang, Hong and Liu, Zimo and Li, Chen},
  journal={arXiv preprint arXiv:2411.14951},
  year={2024}
}

Also consider citing these prior works that are used in this project:

@article{luo2024m,
  title={M $\^{} 3$ GPT: An Advanced Multimodal, Multitask Framework for Motion Comprehension and Generation},
  author={Luo, Mingshuang and Hou, Ruibing and Li, Zhuo and Chang, Hong and Liu, Zimo and Wang, Yaowei and Shan, Shiguang},
  journal={Advances in Neural Information Processing Systems},
  volume={37},
  pages={28051--28077},
  year={2024}
}