morphosymm-rl is a reinforcement learning library for IsaacLab that extends the Proximal Policy Optimization (PPO) implementation of RSL-RL with Morphological Symmetries. This enables data augmentation of robotics systems, or directly embedding symmetry constraints through equivariant neural networks in the actor-critic structure. See the end of this README if you want to cite our work.
Install this package with:
pip install -e .For now, you need to install MorphoSymm and rsl-rl independently. Enter in their folders (they are submodules in this project) and do pip install -e .
To use it, you just need first to change the usual rsl_rl_ppo_cfg.py file of your task in IsaacLab. Second, you need to use ad-hoc train.py and play.py files to launch your training/testing. See here for an example.
If you find the work useful, please consider citing one of our works:
@inproceedings{suhuang2024leveraging,
author={Su, Zhi and Huang, Xiaoyu and Ordoñez-Apraez, Daniel and Li, Yunfei and Li, Zhongyu and Liao, Qiayuan and Turrisi, Giulio and Pontil, Massimiliano and Semini, Claudio and Wu, Yi and Sreenath, Koushil},
booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
title={Leveraging Symmetry in RL-based Legged Locomotion Control},
year={2024},
pages={6899-6906},
doi={10.1109/IROS58592.2024.10802439}
}
@article{ordonez2025morphosymm,
author = {Daniel Ordoñez-Apraez and Giulio Turrisi and Vladimir Kostic and Mario Martin and Antonio Agudo and Francesc Moreno-Noguer and Massimiliano Pontil and Claudio Semini and Carlos Mastalli},
title ={Morphological symmetries in robotics},
journal = {The International Journal of Robotics Research},
year = {2025},
doi = {10.1177/02783649241282422},
URL = {https://doi.org/10.1177/02783649241282422},
eprint = {https://doi.org/10.1177/02783649241282422}
}