UNITREE_REAL.md

Unitree Robot Motion Data Collection

This project provides tools for collecting and processing motion data from Unitree humanoid robots (G1 and H1-2). It supports loading motion sequences from TXT files and collecting real-time robot state data during motion execution using ROS2.

Installation

1. Create and activate a Conda environment:

   conda create -n unitree python=3.9
   conda activate unitree

2. Install ROS2 Foxy: Follow the official ROS2 Foxy installation guide

3. Install Unitree SDK: Refer to Unitree_ros2 for detailed installation instructions

4. Install required Python packages:

   pip install numpy scipy joblib

Usage

Execute Motion and Collect Data

For G1 robot:

python scripts/run_real.py \
    --robot-name g1 \
    --motion-files 'path/to/motion.txt'\
    --output-folder 'path/to/output'

For H1-2 robot:

python scripts/run_real.py \
    --robot-name h12 \
    --motion-files 'path/to/motion.txt'\
    --output-folder 'path/to/output'

Input Format

TXT Motion Files

Motion files should be in CSV format:

joint1,joint2,joint3,joint4,joint5,joint6,...
0.1,0.2,0.3,0.4,0.5,0.6,...
0.2,0.3,0.4,0.5,0.6,0.7,...
...

• First row: Joint names (varies by robot type)
• Subsequent rows: Joint angles in radians for each motion frame
• Frequency: 50Hz

Output

The system generates CSV files for each motion:

output/
└── motion_name/
    ├── joint_list.txt      # List of joint names
    ├── control.csv         # Command data sent to robot
    ├── event.csv          # Motion events and timestamps
    └── state_motor.csv    # Actual robot state data

Robot Configuration

Supported Robots:

• G1: Unitree G1 humanoid robot
• H1-2: Unitree H1-2 humanoid robot

Default Settings:

• Control Frequency: 200Hz
• Motion Frequency: 50Hz
• ROS2 Topics: /lowcmd (command), /lowstate (feedback)