This repository uses ROS2 to implement the entire Sim-to-sim and Sim-to-real workflow. Therefore, ROS2 must first be installed on your computer, such as installing ROS2 Humble on Ubuntu 22.04. We've also released an introduction video, please check it out! Please go through the whole process on a Ubuntu system.
graph LR
A["/rl_deploy"] -->|/JOINTS_CMD| B["/mujoco_simulation"]
B -->|/IMU_DATA| A
B -->|/JOINTS_DATA| A
The following listings show the low-level ROS 2 communication; control commands are described in the control interfaces section below.
# ros2 topic list
/BATTERY_DATA
/IMU_DATA
/JOINTS_CMD
/JOINTS_DATA
/parameter_events
/rosout
# ros2 node info /mujoco_simulation
/mujoco_simulation
Subscribers:
/JOINTS_CMD: drdds/msg/JointsDataCmd
Publishers:
/IMU_DATA: drdds/msg/ImuData
/JOINTS_DATA: drdds/msg/JointsData
/parameter_events: rcl_interfaces/msg/ParameterEvent
/rosout: rcl_interfaces/msg/Log
Service Servers:
/mujoco_simulation/describe_parameters: rcl_interfaces/srv/DescribeParameters
/mujoco_simulation/get_parameter_types: rcl_interfaces/srv/GetParameterTypes
/mujoco_simulation/get_parameters: rcl_interfaces/srv/GetParameters
/mujoco_simulation/list_parameters: rcl_interfaces/srv/ListParameters
/mujoco_simulation/set_parameters: rcl_interfaces/srv/SetParameters
/mujoco_simulation/set_parameters_atomically: rcl_interfaces/srv/SetParametersAtomically
Service Clients:
Action Servers:
Action Clients:
# ros2 node info /rl_deploy
/rl_deploy
Subscribers:
/BATTERY_DATA: drdds/msg/BatteryData
/IMU_DATA: drdds/msg/ImuData
/JOINTS_DATA: drdds/msg/JointsData
/parameter_events: rcl_interfaces/msg/ParameterEvent
Publishers:
/JOINTS_CMD: drdds/msg/JointsDataCmd
/parameter_events: rcl_interfaces/msg/ParameterEvent
/rosout: rcl_interfaces/msg/Log
Service Servers:
/rl_deploy/describe_parameters: rcl_interfaces/srv/DescribeParameters
/rl_deploy/get_parameter_types: rcl_interfaces/srv/GetParameterTypes
/rl_deploy/get_parameters: rcl_interfaces/srv/GetParameters
/rl_deploy/list_parameters: rcl_interfaces/srv/ListParameters
/rl_deploy/set_parameters: rcl_interfaces/srv/SetParameters
/rl_deploy/set_parameters_atomically: rcl_interfaces/srv/SetParametersAtomically
Service Clients:
Action Servers:
Action Clients:
Everyone is welcome to contribute to this repo. If you discover a bug or optimize our training config, just submit a pull request and we will look into it.
pip install "numpy < 2.0" mujoco scipy
git clone https://github.com/DeepRoboticsLab/sdk_deploy.git
# Compile
cd sdk_deploy
source /opt/ros/<ros-distro>/setup.bash
colcon build --packages-up-to m20_sdk_deploy --cmake-args -DBUILD_PLATFORM=x86Start the simulator in Terminal 2 before starting the controller in Terminal 1. For simultaneous robots, use separate DDS domains for all their joint and IMU topics.
# Run (Open 2 terminals)
# Terminal 1
export ROS_DOMAIN_ID=1
source install/setup.bash
ros2 run m20_sdk_deploy rl_deploy
# Terminal 2
export ROS_DOMAIN_ID=1
source install/setup.bash
python3 src/M20_sdk_deploy/interface/robot/simulation/mujoco_simulation_ros2.pyFor Sim-to-sim tests, use only the keyboard or ROS 2 interface. Refer to Control interfaces for selection instructions and commands.
Note:
- Right click simulator window and select "always on top"
- When the robot dog stands up, it may become stuck due to self-collision in the simulation. This is not a bug; please try again.
This process is almost identical to simulation-simulation. You only need to add the step of connecting to Wi-Fi to transfer data, and then modify the compilation instructions.Real-robot control supports keyboard, gamepad, and ROS 2 input. Select the desired RemoteCommandType in main.cpp and rebuild; the default is kKeyBoard.
Please first use the OTA upgrade function in the handle settings to upgrade the hardware to version 1.1.8. We require a sdk authentication code to activate the sdk mode. Please contact our technical support team to get this unique code for each robot.
# computer and gamepad should both connect to WiFi
# WiFi: CA9B********
# Passward: 12345678 (If wrong, contact technical support)
# Note: If you are connected via the second WiFi, use 10.21.41.1 instead of 10.21.31.103
# for ssh/scp below. It is the same robot, only the IP differs per WiFi network.
# scp to transfer files to M20 (open a terminal on your local computer)
# Include the updated drdds source to generate RobotStatus in the workspace.
ssh user@10.21.31.103 'mkdir -p ~/M20_sdk_deploy/src'
scp -r ~/sdk_deploy/src/drdds ~/sdk_deploy/src/M20_sdk_deploy user@10.21.31.103:~/M20_sdk_deploy/src
# ssh connect for remote development,
ssh user@10.21.31.103
cd M20_sdk_deploy
source /opt/ros/foxy/setup.bash #source ROS2 env
# Build drdds and the SDK into this workspace; keep the firmware installation
# under /opt/ros/foxy unchanged. Source this workspace when running the new SDK.
colcon build --packages-up-to m20_sdk_deploy --cmake-args -DBUILD_PLATFORM=arm
sudo su # Root
source /opt/ros/foxy/setup.bash #source ROS2 env
source /opt/robot/scripts/setup_ros2.sh
# Use the gamepad to enable SDK mode. Need authorization code, please contact technical support team.
# Run
source install/setup.bash
ros2 run m20_sdk_deploy rl_deploy
# exit sdk mode：
# Use the gamepad to enable SDK mode.Use the build and deployment steps above before selecting an input method; the
default selection in main.cpp is kKeyBoard.
To use keyboard control, set RemoteCommandType::kKeyBoard (0) in main.cpp, rebuild, and restart the SDK.
Press the following keys in the terminal running rl_deploy:
- z： default position
- c： rl control default position
- x： lie down
- wasd：forward/leftward/backward/rightward
- qe：counter clockwise/clockwise
To use gamepad control, set RemoteCommandType::kGamepad (1) in main.cpp, rebuild, and restart the SDK.
(Note: When using the gamepad control function, please ensure that the Gamepad APP version is V1.5.11 or higher.)
- L1： default position
- L2： rl control default position
- R1： lie down
- R2： joint damping
- Left joystick：forward/leftward/backward/rightward
- Right joystick：clockwise/counter clockwise
To use ROS 2 control, set RemoteCommandType::kRos2 (2) in main.cpp, rebuild, and restart the SDK.
Start the controller with ros2 run m20_sdk_deploy rl_deploy; in another
terminal, source the same ROS/workspace environment and use the same DDS domain.
For the Sim-to-sim setup above, run export ROS_DOMAIN_ID=1 in this command terminal too.
Stand up:
ros2 topic pub --once /rl_deploy/command_state std_msgs/msg/String '{data: stand}'Enter policy control after standing has finished:
ros2 topic pub --once /rl_deploy/command_state std_msgs/msg/String '{data: rl_control}'Simulation movement example (forward at a requested 0.1 m/s):
ros2 topic pub --rate 50 /rl_deploy/cmd_vel geometry_msgs/msg/Twist \
'{linear: {x: 0.1, y: 0.0}, angular: {z: 0.0}}'Use linear.x for forward/backward (m/s), linear.y for left/right (m/s), and
angular.z for counterclockwise/clockwise yaw (rad/s); other axes must be zero.
Nonzero velocity is used only in policy control. Press Ctrl+C to stop publishing;
after 0.25 seconds without velocity input, the effective command becomes zero.
Lie down:
ros2 topic pub --once /rl_deploy/command_state std_msgs/msg/String '{data: lie_down}'Request joint damping:
ros2 topic pub --once /rl_deploy/command_state std_msgs/msg/String '{data: damping}'Read current state, commands, joint data, and base orientation/angular velocity:
ros2 topic echo /rl_deploy/status
# In a separate terminal, check the effective command rate:
ros2 topic hz /rl_deploy/effective_cmd_vel/rl_deploy/status publishes drdds/msg/RobotStatus
at 10 Hz; effective commands are published at 50 Hz. Default velocity limits are
0.7 m/s forward, 0.5 m/s sideways, and 0.7 rad/s yaw. For posture-only operation,
start the controller with all three limits at zero:
ros2 run m20_sdk_deploy rl_deploy --ros-args \
-p max_forward_velocity:=0.0 -p max_side_velocity:=0.0 -p max_yaw_velocity:=0.0Simulation command and stand/lie checks passed, but the existing policy shows negligible sideways motion, yaw overshoot, and drift at zero input. Hardware checks covered stand/lie only.