Unitree Go2 / Go2W / B2 — Edge Deployment Guide

简体中文文档: 点此查看

Overview

This guide documents the practical steps to deploy perception and motion-control components on Unitree robots (Go2 / Go2W / B2) using the InternNav project (https://github.com/InternRobotics/InternNav). The focus is on edge-side (robot-side) deployment and debugging. For server-side setup and model serving, consult the InternNav documentation.

Compatibility and scope:

• Target hardware: Unitree Go2 (onboard compute box) and compatible edge devices
• Recommended OS: Ubuntu 20.04 LTS (tested with ROS 2 Foxy)
• Python: 3.8+
• Camera: Intel RealSense series (recommended D455)

---

1. Hardware Preparation

1. Update the robot firmware to the latest version to ensure compatibility and stability.
2. Set the robot's motion mode to AI mode via the Unitree mobile app — this reduces torso motion and improves visual stability.
3. Connection options:
   • Onboard compute box: uses the robot's internal network (no additional hardware required).
   • External compute (NUC / Jetson): connect via Ethernet. Default robot subnet: 192.168.123.x. See Unitree SDK2 docs for interface setup.

1.1 Camera mounting

• Recommended model: Intel RealSense D455 (global shutter reduces motion blur).
• Use the provided 3D-print mounts for better forward ground view: https://github.com/InternRobotics/InternNav/tree/main/assets/3d_printing_files
• Use a USB 3.2 Gen 1 (or better) data cable and plug into the front-panel Type-C port of the compute box to avoid bandwidth issues.

---

2. Software Environment

2.1 Install Unitree SDK2 (Python)

cd ~
sudo apt update
sudo apt install -y python3-pip git
git clone https://github.com/unitreerobotics/unitree_sdk2_python.git
cd unitree_sdk2_python
pip3 install -e .

2.2 Intel RealSense

Install the librealsense system packages and the Python bindings. See the official docs for Jetson-specific steps.

# Add RealSense key and repo (example)
sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-key F6E65AC044F831AC80A06380C8B3A55A6F3EFCDE
sudo add-apt-repository "deb https://librealsense.intel.com/Debian/apt-repo $(lsb_release -cs) main" -u
sudo apt-get install -y librealsense2-utils librealsense2-dev

# Python binding
pip3 install pyrealsense2

2.3 InternNav client environment

git clone https://github.com/InternRobotics/InternNav.git
cd InternNav
pip3 install numpy requests Pillow casadi

2.4 CycloneDDS (if required)

If you encounter "Could not locate cyclonedds" when installing SDKs, compile CycloneDDS and set CYCLONEDDS_HOME:

cd ~
git clone https://github.com/eclipse-cyclonedds/cyclonedds -b releases/0.10.x
cd cyclonedds
mkdir build install && cd build
cmake .. -DCMAKE_INSTALL_PREFIX=../install
cmake --build . --target install

# Use absolute path on export
export CYCLONEDDS_HOME="/home/<user>/cyclonedds/install"
cd ~/unitree_sdk2_python
pip3 install -e .

---

3. Repository Layout (onboard)

Key modules in InternNav/onboard:

• start_robot.sh — single script to start driver, ROS 2 bridge, and camera node; includes simple supervision and configurable network interface.
• unitree_driver.py — low-level UDP driver (port 8899) wrapping Unitree SDK calls.
• ros_bridge.py — ROS 2 bridge publishing /odom_bridge and forwarding /cmd_vel_bridge to the driver.
• realsense_publisher.py — publishes RGB and aligned depth (640×480 @30fps) and implements auto-reconnect logic.

---

4. Start-up & Quick Start

1. Make the launcher executable:

cd onboard
chmod +x start_robot.sh

2. Edit network interface in start_robot.sh if your interface is not eth0:

ROBOT_IF="eth0"  # change to enp3s0, wlan0, etc. as needed

3. Start the system:

./start_robot.sh

4. Configure and start the navigation client:

Configure the inference server address:

cd InternNav
nano scripts/realworld/http_internvla_client.py

Find the following line and replace 127.0.0.1 with the actual server IP:

url='http://127.0.0.1:5801/eval_dual'  # replace 127.0.0.1 with the inference server IP

Then start the robot-side modules:

Terminal 1 (robot):

cd InternNav/onboard
./start_robot.sh

Terminal 2 (onboard or remote client):

cd InternNav
python3 scripts/realworld/http_internvla_client.py

If running a local inference server, on the server machine:

python3 scripts/realworld/http_internvla_server.py

---

5. ROS 2 Topics & Visualization

Subscribe with RViz2 to validate topics:

Topic	Type	Description
/camera/camera/color/image_raw	sensor_msgs/Image	RGB image
/camera/camera/aligned_depth_to_color/image_raw	sensor_msgs/Image	Aligned depth image
/odom_bridge	nav_msgs/Odometry	Robot odometry

Example RViz2 visualization:

---

6. Troubleshooting

• Camera not detected: check USB cable (USB 3.2 Gen1), ensure front Type-C port, run rs-enumerate-devices.
• Network issues: verify interface with ifconfig / ip addr, ping 192.168.123.161, and ensure UDP 8899 is reachable.
• ROS 2 no topics: check echo $ROS_DOMAIN_ID and ensure same ROS_DOMAIN_ID across devices; ros2 topic list to inspect.

---

7. References

• InternNav docs: https://internrobotics.github.io/user_guide/internnav/
• InternNav repo: https://github.com/InternRobotics/InternNav/
• Unitree SDK2: https://github.com/unitreerobotics/unitree_sdk2
• Intel RealSense: https://github.com/IntelRealSense/librealsense