RoboMimic Deploy

🌎English | 🇨🇳中文

​RoboMimic Deploy​​ is a multi-policy robot deployment framework based on a state-switching mechanism. Currently, the included policies are designed for the ​​Unitree G1 robot (29-DoF)​​.

Preface

• ​This deployment framework is only applicable to G1 robots with a 3-DOF waist. If a waist fixing bracket is installed, it must be unlocked according to the official tutorial before this framework can be used normally.​​

• It is recommended to remove the hands, as dance movements may cause interference.​

• video instruction

Installation and Configuration

1. Create a Virtual Environment

It is recommended to run training or deployment programs in a virtual environment. We suggest using Conda to create one.

1.1 Create a New Environment

Use the following command to create a virtual environment:

conda create -n robomimic python=3.8

1.2 Activate the Virtual Environment

conda activate robomimic

---

2. Install Dependencies

2.1 Install PyTorch

PyTorch is a neural network computation framework used for model training and inference. Install it with the following command:

conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 pytorch-cuda=12.1 -c pytorch -c nvidia

2.2 Install RoboMimic_Deploy

2.2.1 Download

Clone the repository via git:

git clone https://github.com/ccrpRepo/RoboMimic_Deploy.git

2.2.2 Install Components

Navigate to the directory and install:

cd RoboMimic_Deploy
pip install numpy==1.20.0
pip install onnx onnxruntime

2.2.3 Install unitree_sdk2_python

git clone https://github.com/unitreerobotics/unitree_sdk2_python.git
cd unitree_sdk2_python
pip install -e .

---

Running the Code

1. Run Mujoco Simulation

python deploy_mujoco/deploy_mujoco.py

2. Policy Descriptions

Mode Name	Description
PassiveMode	Damping protection mode
FixedPose	Position control reset to default joint values
LocoMode	Stable walking control mode
Dance	Charleston dance routine
KungFu	Martial arts movement
KungFu2	Failed martial arts training
Kick	Bad mimic policy
SkillCast	Lower body + waist stabilization with upper limbs positioned to specific joint angles (typically executed before Mimic strategy)
SkillCooldown	Lower body + waist continuous balancing with upper limbs reset to default angles (typically executed after Mimic strategy)

---

3. Operation Instructions in Simulation

1. Connect an Xbox controller.
2. Run the simulation program:

python deploy_mujoco/deploy_mujoco.py

3. Press the ​​Start​​ button to enter position control mode.

4. Hold ​​R1 + A​​ to enter ​​LocoMode​​, then press BACKSPACE in the simulation to make the robot stand. Afterward, use the joystick to control walking.

5. Hold ​​R1 + X​​ to enter ​​Dance​​ mode—the robot will perform the Charleston. In this mode:

   • Press ​​Select​​ at any time to switch to damping protection mode.
   • Hold ​​R1 + A​​ to return to walking mode (not recommended).
   • Press ​​Start​​ to return to position control mode.

6. The terminal will display a progress bar for the dance. After completion, press ​​R1 + A​​ to return to normal walking mode.

7. In ​​LocoMode​​, pressing ​​R1 + Y​​ triggers a Martial arts movement —​ ​use only in simulation​​.

8. In ​​LocoMode​​, pressing ​​L1 + Y​​ triggers a Martial arts movement(Failed) —​ ​use only in simulation​​.

9. In ​​LocoMode​​, pressing ​​R1 + B​ triggers a Kick movement(Failed) —​ ​use only in simulation​​.

---

4. Real Robot Operation Instructions

1. Power on the robot and suspend it (e.g., with a harness). and then hold L2+R2

2. Run the deploy_real program:

python deploy_real/deploy_real.py

3. Press the ​​Start​​ button to enter position control mode.
4. Subsequent operations are the same as in simulation.

---

Important Notes

1. Framework Compatibility Notice

The current framework does not natively support deployment on G1 robots equipped with Orin NX platforms. Preliminary analysis suggests compatibility issues with the unitree_python_sdk on Orin systems. For onboard Orin deployment, we recommend the following alternative solution:

• Replace with unitree_sdk2 (official C++ SDK)
• Implement a dual-node ROS architecture:
  • C++ Node: Handles data transmission between robot and controller
  • Python Node: Dedicated to policy inference

2. Mimic Policy Reliability Warning

The Mimic policy does not guarantee 100% success rate, particularly on slippery/sandy surfaces. In case of robot instability:

• Press F1 to activate PassiveMode (damping protection)
• Press Select to immediately terminate the control program

3. Charleston Dance (R1+X) - Stable Policy Notes

Currently the only verified stable policy on physical robots:

⚠️ Important Precautions:

• Palm Removal Recommended: The original training didn't account for palm collisions (author's G1 lacked palms)
• Initial/Final Stabilization: Brief manual stabilization may be required when starting/ending the dance
• Post-Dance Transition: While switching to Locomotion/PositionControl/PassiveMode is possible, we recommend:
  • First transition to PositionControl or PassiveMode
  • Provide manual stabilization during transition

4. Other Movement Advisories

All other movements are currently not recommended for physical robot deployment.

5. Strong Recommendation

Always master operations in simulation before attempting physical robot deployment.