Training and Play

Wiki-GRx-Gym provides a training environment based on NVIDIA Isaac Gym, combined with the legged_gym and rsl_rl libraries from the Legged Robotics team at ETH Zurich, for training the Fourier N1 robot's walking ability on complex terrain.

Related Resources

• NVIDIA Isaac Gym: https://developer.nvidia.com/isaac-gym
• legged_gym: https://github.com/leggedrobotics/legged_gym.git
• rsl_rl: https://github.com/leggedrobotics/rsl_rl.git

Installation Guide

1. Install Ubuntu 20.04 / Ubuntu 22.04 system

2. Conda Environment Configuration

   # Install Miniconda
   cd ~/Downloads
   wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
   bash Miniconda3-latest-Linux-x86_64.sh
   
   # Create training environment
   conda create -n wiki-grx-gym python=3.8 -y
   conda activate wiki-grx-gym

3. Dependency Installation

   # Navigate to the project directory
   cd path/to/your/project
   
   # Install Isaac Gym
   cd IsaacGym_Preview_4_Package/isaacgym/python/
   pip install -e . -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
   
   # Navigate to the project directory
   cd path/to/your/project
   
   # Install rsl_rl
   cd rsl_rl
   pip install -e . -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
   
   # Navigate to the project directory
   cd path/to/your/project
   
   # Install legged_gym
   cd legged_gym 
   pip install -e . `-i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple`
   
   # Install other dependencies
   pip install tensorboard protobuf==3.20.3 -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple

Usage Instructions

1. Start Training

   cd legged_gym/legged_gym/scripts
   python train.py --task=N1 --headless

2. Demonstration Test

   python play.py --task=N1 --num_envs=1

3. Export Policy

   • When running play.py, the policy network model will be automatically exported to logs/N1/exported/policy_jit.pt.
   • This policy model can be used for subsequent deployment on the real robot.

[!NOTE]

The average training time using NVIDIA RTX 4090 is about 6 seconds per iteration. Completing 5000 iterations takes approximately 8 hours. Actual training time can be adjusted based on reward growth and training objectives.

Common Issues

1. Ubuntu 22.04 reports error "libpython3.8.so.1.0: cannot open shared object file"

   • The displayed information is: ImportError: libpython3.8.so.1.0: cannot open shared object file: No such file or directory
   • Solution reference: https://blog.csdn.net/weixin_43989965/article/details/136612205
   • You can try activating the wiki-grx-gym environment first, running the automatic configuration script in this project, then exiting the conda environment, and then reactivating the corresponding conda environment:

   conda activate wiki-grx-gym
   bash shell/conda_import_libpython.sh
   conda deactivate
   conda activate wiki-grx-gym