🦿 Legged Robotics in Genesis

A legged_gym based framework for training legged robots in genesis

Table of Contents

• 🦿 Legged Robotics in Genesis
  • Table of Contents
  • 📅 Updates
  • 🌟 Features
  • 🧪 Test Results
  • 🛠 Installation
  • 👋 Usage
    • 🚀 Quick Start
    • 📖 Instructions
  • 🖼️ Gallery
  • 🙏 Acknowledgements
  • TODO

---

📅 Updates

2025/02/23

• find some bugs in genesis. The environments can return nan values without knowing why (Genesis-Embodied-AI/Genesis#625), which can hinder long-time training where nan values may suddenly come out and terminates your training.

  

2025/02/10

• add measure_heights support, and provide a demo of exteroceptive locomotion (go2_rough)

2024/12/28

• add wiki page for instructions

2024/12/26

• add terrain support, optional terrain type: ["plane", "heightfield"].

  

• move test results to tests.md

2024/12/24

• add a new demo environment bipedal_walker

2024/12/23

• divide main and deploy branches, deploy branch should be used with a custom rsl_rl(which will be open-source soon)

---

🌟 Features

• Totally based on legged_gym

  It's easy to use for those who are familiar with legged_gym and rsl_rl

• Faster and Smaller

  For a go2 walking on the plane task with 4096 envs, the training speed in Genesis is approximately 1.3x compared to Isaac Gym, while the graphics memory usage is roughly 1/2 compared to IsaacGym.

  With this smaller memory usage, it's possible to run more parallel environments, which can further improve the training speed.

🧪 Test Results

For tests conducted on Genesis, please refer to tests.md

🛠 Installation

1. Create a new python virtual env with python>=3.9
2. Install PyTorch
3. Install Genesis following the instructions in the Genesis repo
4. Install rsl_rl and tensorboard

   # Install rsl_rl.
   git clone https://github.com/leggedrobotics/rsl_rl
   cd rsl_rl && git checkout v1.0.2 && pip install -e . --use-pep517
   
   # Install tensorboard.
   pip install tensorboard
5. Install genesis_lr

   git clone https://github.com/lupinjia/genesis_lr
   cd genesis_lr
   pip install -e .

👋 Usage

🚀 Quick Start

By default, the task is set to go2(in utils/helper.py), we can run a training session with the following command:

cd legged_gym/scripts
python train.py --headless # run training without rendering

After the training is done, paste the run_name under logs/go2 to load_run in go2_config.py:

Then, run play.py to visualize the trained model:

📖 Instructions

For more detailed instructions, please refer to the wiki page

🖼️ Gallery

Go2	Bipedal Walker

🙏 Acknowledgements

• Genesis
• Genesis-backflip
• legged_gym
• rsl_rl
• unitree_rl_gym

TODO

• Add domain randomization
• Verify the trained model on real robots.
• Add Heightfield support
• Add meausre_heights support
• Add go2 deploy demos and instructions (vanilla and explicit estimator)
• Add teacher-student implementation