Legged RL Deploy

A deployment framework for reinforcement learning (RL) policies on Unitree Go2 and G1 robots.

Features

• Configurable via config.yaml
• Support for TorchScript and ONNX Runtime inference
• Multiple pretrained policies included

Policy sources

policies/
├── go2/
│   ├── velocity/
│   │   ├── unitree_rl_lab/    # unitree_rl_lab
│   │   ├── mjlab/             # mjlab
│   │   └── aba/               # mjlab
│   ├── hop/                   # My_unitree_go2_gym
│   ├── trot/                  # My_unitree_go2_gym
│   ├── handstand/             # My_unitree_go2_gym
│   ├── legstand/              # My_unitree_go2_gym
│   └── spring_jump/           # My_unitree_go2_gym
└── g1/
    ├── velocity/
    │   └── unitree_rl_mjlab/  # unitree_rl_mjlab
    └── mimic/
        ├── gangnam_style/     # unitree_rl_lab
        ├── dance_102/         # unitree_rl_lab
        ├── dance1_subject2/   # unitree_rl_mjlab
        └── pose/              # whole_body_tracking
        └── LAFAN1_Retargeting # whole_body_tracking

Source	Link
unitree_rl_lab	github.com/unitreerobotics/unitree_rl_lab
My_unitree_go2_gym	github.com/yusongmin1/My_unitree_go2_gym
mjlab	github.com/mujocolab/mjlab
unitree_rl_mjlab	github.com/mujocolab/unitree_rl_mjlab

Thanks to the authors of these projects for their contributions.

Video Demonstrations

• Go2 Hop (Bilibili)

Installation

Dependencies

• LibTorch

• ORT

• unitree_lowlevel

Build

cd unitree_ws/src
git clone https://github.com/Renkunzhao/legged_rl_deploy.git
cd ..

# Ensure CMAKE_PREFIX_PATH and LD_LIBRARY_PATH include torch paths 
source install/setup.bash
colcon build --packages-up-to legged_rl_deploy 

Run

# Run test and stop the default controller (run once per boot)
source src/unitree_lowlevel/scripts/setup.sh <network-interface> $ROS_DISTRO
./build/unitree_sdk2/bin/go2_stand_example $NetworkInterface

source src/unitree_lowlevel/scripts/setup.sh <network-interface> $ROS_DISTRO
ros2 run legged_rl_deploy legged_rl_deploy_node $NetworkInterface $WORKSPACE/src/legged_rl_deploy/config/go2-trot.yaml

Troubleshooting

CMake CUDA Dialect Error:

Target "cmTC_dfc1b" requires the language dialect "CUDA17" (with compiler extensions), 
but CMake does not know the compile flags to use to enable it.

Solution: Update CMake to a newer version that supports CUDA17.