We are happy to present you our integration of the Unitree Go2 with ROS2 over Wi-Fi, that was designed by the talented @tfoldi. You can explore his groundbreaking work at go2-webrtc.
This repo will empower your Unitree GO2 AIR/PRO/EDU robots with ROS2 capabilities, using both WebRTC (Wi-Fi) and CycloneDDS (Ethernet) protocols.
If you are using WebRTC (Wi-Fi) protocol, close the connection with a mobile app before connecting to the robot.
If you're as enthusiastic about this project as we are, please consider giving it a :star: star!!!
Your encouragement fuels our passion and helps us develop our RoadMap further. We welcome any help or suggestions you can offer!
Together, let's push the boundaries of what's possible with the Unitree Go2 and ROS2!
:sparkles: Full ROS2 SDK support for your Unitree GO2
:robot: Compatible with AIR, PRO, and EDU variants
:footprints: Access to foot force sensors feedback (available on some GO2 PRO models or EDU)
Tested systems and ROS2 distro | systems | ROS2 distro | Build status |
---|---|---|---|
Ubuntu 22.04 | iron | ||
Ubuntu 22.04 | humble | ||
Ubuntu 22.04 | rolling |
mkdir -p ros2_ws
cd ros2_ws
git clone --recurse-submodules https://github.com/abizovnuralem/go2_ros2_sdk.git src
sudo apt install ros-$ROS_DISTRO-image-tools
sudo apt install ros-$ROS_DISTRO-vision-msgs
sudo apt install python3-pip clang portaudio19-dev
cd src
pip install -r requirements.txt
cd ..
Pay attention to any error messages. If pip install
does not complete cleanly, various features will not work. For example, open3d
does not yet support python3.12
and therefore you will need to set up a 3.11 venv
first etc.
Build go2_ros_sdk
. You need to have ros2
and rosdep
installed. If you do not, follow these instructions. Then:
source /opt/ros/$ROS_DISTRO/setup.bash
rosdep install --from-paths src --ignore-src -r -y
colcon build
Don't forget to set up your Go2 robot in Wifi-mode and obtain the IP. You can use the mobile app to get it. Go to Device -> Data -> Automatic Machine Inspection and look for STA Network: wlan0.
source install/setup.bash
export ROBOT_IP="robot_ip" #for muliple robots, just split by ,
export CONN_TYPE="webrtc"
ros2 launch go2_robot_sdk robot.launch.py
The robot.launch.py
code starts many services/nodes simultaneously, including
joy
(ROS2 Driver for Generic Joysticks and Game Controllers)teleop_twist_joy
(facility for tele-operating Twist-based ROS2 robots with a standard joystick. Converts joy messages to velocity commands) twist_mux
(twist_multiplexer with source prioritization) When you run robot.launch.py
, rviz
will fire up, lidar data will begin to accumulate, the front color camera data will be displayed too (typically after 4 seconds), and your dog will be waiting for commands from your joystick (e.g. a X-box controller). You can then steer the dog through your house, e.g., and collect LIDAR mapping data.
The goal of SLAM overall, and the slam_toolbox
in particular, is to create a map. The slam_toolbox
is a grid mapper - it thinks about the world in terms of a fixed grid that the dog operates in. When the dog initially moves through a new space, data accumulate and the developing map is and published it to the /map
topic. The goal of Nav2
is to navigate and perform other tasks in this map.
The rviz
settings that are used upon initial launch (triggered by ros2 launch go2_robot_sdk robot.launch.py
) showcase various datastreams.
RobotModel
is the dimensionally correct model of the G02 PointCloud2
are the raw LIDAR data transformed into 3D objects/constraints LaserScan
are lower level scan data before translation into an x,y,z frameImage
are the data from the front-facing color camera Map
is the map being created by the slam_toolbox
Odometry
is the history of directions/movements of the dogIf there is too much going on in the initial screen, deselect the map
topic to allow you to see more.
Use painter's tape to mark a 'dock' rectangle (or use a real dock) to create a defined starting point for your dog on your floor. In the rviz
SlamToolboxPlugin
, on the left side of the your rviz
screen, select "Start At Dock". Then, use your controller to manually explore a space, such as a series of rooms. You will see the map data accumulating in rviz
. In this map, white, black and grey pixels represent the free, occupied, and unknown space, respectively. When you are done mapping, enter a file name into the "Save Map" field and click "Save Map". Then enter a file name into "Serialize Map" field and click "Serialize Map". Now, you should have 2 new files in /ros2_ws
:
map_1.yaml: the metadata for the map as well as the path to the .pgm image file.
map_1.pgm: the image file with white, black and grey pixels representing the free, occupied, and unknown space.
map_1.data:
map_1.posegraph:
The next time you start the system, the map can be loaded and is ready for you to complete/extend by mapping more spaces. Upon restart and loading a map, the dog does not know where it is relative to the map you created earlier. Assuming you rebooted the dog in its marked rectangle, or in an actual dock, it will have a high quality initial position and angle.
As shown in the rviz
Navigation 2
plugin, the system will come up in:
Navigation: active
Localization: inactive
Feedback: unknown
Then, load your map via the SlamToolboxPlugin
(enter your map's filename (without any extension) in the 'Deserialize Map' field and then click 'Deserialize Map').
WARNING: please make sure that (1) the dog is correctly oriented WRT to the map and (2) the map itself is sane and corresponds to your house. Especially if you have long corridors, the overall map can be distorted relative to reality, and this means that the route planner will try to route your dog through walls, leaving long scratches in your walls.
You can now give the dog its first target, via 'Nav2 Goal' in the rviz
menu. Use the mouse cursor to provide a target to navigate to.
NOTE: the Nav2 Goal
cursor sets both the target position and the final angle of the dog, that you wish the dog to adopt upon reaching the target (need to double check). The long green arrow that is revealed when you click an point and keep moving your mouse cursor is the angle setter.
Until you have some experience, we suggest following your dog and picking it up when it is about to do something silly.
NOTE: Virtually all fault behaviors - spinning in circles, running into walls, trying to walk through walls, etc reflect (1) a map that is incorrect, (2) incorrect initial position/angle of the dog relative to that map, or (3) inability to compute solutions/paths based on overloaded control loops. To prevent #3, which results in no motion or continuous spinning, the key loop rates (controller_frequency
: 3.0 and expected_planner_frequency
: 1.0 have been set to very conservative rates).
This capability is directly based on J. Francis's work. Launch the go2_ro2_sdk
. After a few seconds, the color image data will be available at go2_camera/color/image
. On another terminal enter:
source install/setup.bash
ros2 run coco_detector coco_detector_node
There will be a short delay the first time the node is run for PyTorch TorchVision to download the neural network. You should see a download progress bar. TorchVision cached for subsequent runs.
On another terminal, to view the detection messages:
source install/setup.bash
ros2 topic echo /detected_objects
The detection messages contain the detected object (class_id
) and the score
, a number from 0 to 1. For example: detections:results:hypothesis:class_id: giraffe
and detections:results:hypothesis:score: 0.9989
. The bbox:center:x
and bbox:center:y
contain the centroid of the object in pixels. These data can be used to implement real-time object following for animals and people. People are detected as detections:results:hypothesis:class_id: person
.
To view the image stream annotated with the labels and bounding boxes:
source install/setup.bash
ros2 run image_tools showimage --ros-args -r /image:=/annotated_image
Example Use:
ros2 run coco_detector coco_detector_node --ros-args -p publish_annotated_image:=False -p device:=cuda -p detection_threshold:=0.7
This will run the coco detector without publishing the annotated image (it is True by default) using the default CUDA device (device=cpu by default). It sets the detection_threshold to 0.7 (it is 0.9 by default). The detection_threshold should be between 0.0 and 1.0; the higher this number the more detections will be rejected. If you have too many false detections try increasing this number. Thus only Detection2DArray messages are published on topic /detected_objects.
To save raw LIDAR data, export
the following:
export MAP_SAVE=True
export MAP_NAME="3d_map"
Every 10 seconds, pointcloud data (in .ply
format) will be saved to the root folder of the repo. NOTE: This is not a Nav2 map but a raw data dump of LIDAR data useful for low-level debugging.
If you want to connect several robots for collaboration:
export ROBOT_IP="robot_ip_1, robot_ip_2, robot_ip_N"
export CONN_TYPE="webrtc"
or
export CONN_TYPE="cyclonedds"
To use Foxglove, you need to install Foxglove Studio:
sudo snap install foxglove-studio
If you are running ROS2 under WSL2 - you may need to configure Joystick\Gamepad to navigate the robot.
Step 1 - share device with WSL2
Follow steps here https://learn.microsoft.com/en-us/windows/wsl/connect-usb to share your console device with WSL2
Step 2 - Enable WSL2 joystick drivers
WSL2 does not come by default with the modules for joysticks. Build WSL2 Kernel with the joystick drivers. Follow the instructions here: https://github.com/dorssel/usbipd-win/wiki/WSL-support#building-your-own-wsl-2-kernel-with-additional-drivers If you're comfortable with WSl2, skip the export steps and start at Install prerequisites.
Before buiding, edit .config
file and update the CONFIG_ values listed in this GitHub issue: https://github.com/microsoft/WSL/issues/7747#issuecomment-1328217406
Step 3 - Give permissions to /dev/input devices
Once you've finished the guides under Step 3 - you should be able to see your joystick device under /dev/input
ls /dev/input
by-id by-path event0 js0
By default /dev/input/event* will only have root permissions, so joy node won't have access to the joystick
Create a file /etc/udev/rules.d/99-userdev-input.rules
with the following content:
KERNEL=="event*", SUBSYSTEM=="input", RUN+="/usr/bin/setfacl -m u:YOURUSERNAME:rw $env{DEVNAME}"
Run as root: udevadm control --reload-rules && udevadm trigger
Step 3 - verify that joy node is able to see the device properly.
Run ros2 run joy joy_enumerate_devices
ID : GUID : GamePad : Mapped : Joystick Device Name
-------------------------------------------------------------------------------
0 : 030000005e040000120b000007050000 : true : false : Xbox Series X Controller
Special thanks to:
This project is licensed under the BSD 2-clause License - see the LICENSE file for details.