This repository integrates reinforcement learning (RL), navigation, and simulation to enable autonomous navigation for the Unitree Go2, Go2W and B2 quadruped robots in Gazebo.
This project follows a Sim-to-Sim approach, executing controllers trained in Isaac Lab within the Gazebo simulator. The virtual Unitree Go2, Go2W and B2 robots are equipped with a LiDAR. The mounting position of the Velodyne VLP16 LiDAR is based on the Unitree developer documentation.
Before you begin, ensure you have the following dependencies installed:
-
ROS 2 Humble: Please follow the official installation instructions.
-
Required ROS 2 Packages:
sudo apt update && sudo apt install -y \ ros-humble-teleop-twist-keyboard \ ros-humble-ros2-control \ ros-humble-ros2-controllers \ ros-humble-control-toolbox \ ros-humble-robot-state-publisher \ ros-humble-joint-state-publisher-gui \ ros-humble-gazebo-ros2-control \ ros-humble-gazebo-ros-pkgs \ ros-humble-xacro \ ros-humble-navigation2 \ ros-humble-nav2-bringup \ ros-humble-octomap-ros\ ros-humble-octomap-rviz-plugins\ ros-humble-velodyne-laserscan -
LibTorch (C++):
# Choose a directory to store the library mkdir -p ~/libs && cd ~/libs wget https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-2.0.1%2Bcpu.zip unzip libtorch-cxx11-abi-shared-with-deps-2.0.1+cpu.zip rm libtorch-cxx11-abi-shared-with-deps-2.0.1+cpu.zip echo 'export Torch_DIR=$HOME/libs/libtorch' >> ~/.bashrc source ~/.bashrc
-
yaml-cpp and lcm:
sudo apt install -y liblcm-dev libyaml-cpp-dev
-
Clone the Repository: Create a ROS 2 workspace and clone this repository recursively.
mkdir -p ~/ros2_ws/src && cd ~/ros2_ws/src git clone https://github.com/RCILab/RCI_quadruped_robot_navigation --recursive
-
Build the Workspace: Install dependencies and build the packages.
cd ~/ros2_ws colcon build --symlink-install
-
Source the Environment: Source the workspace's setup file to make the packages available in your environment.
echo "source ~/ros2_ws/install/setup.bash" >> ~/.bashrc source ~/.bashrc
Follow these steps to launch the simulation and control the robot. Each command should be run in a new terminal.
-
Launch Gazebo & RViz2: This command starts the Gazebo simulation with the robot model and launches RViz2 for visualization.
- Default robot is
go2. - Use the
rnameargument to switch to another robot likego2w.
# Launch with the default Go2 robot ros2 launch rl_sar gazebo.launch.py # Launch with the Go2W robot ros2 launch rl_sar gazebo.launch.py rname:="go2w" # Launch with the Go2W robot ros2 launch rl_sar gazebo.launch.py rname:="b2"
- Default robot is
-
Run the RL Controller: This node is responsible for the robot's locomotion control.
ros2 run rl_sar rl_sim
-
Run the Action Client: This client sends high-level commands or goals to the controller.
ros2 run rl_action command -
Execute Navigation (Nav2) 4.1 Convert Velodyne PointCloud2 -> LaserScan (/scan)
ros2 run velodyne_laserscan velodyne_laserscan_node --ros-args \
-r velodyne_points:=/velodyne_points \
-r scan:=/scan4.2 Launch Nav2
ros2 launch rl_sar nav2.launch.py-
Elevator Control (Optional Feature) This repository includes a Gazebo Elevator Plugin to simulate multi-floor vertical transport. See below for detailed usage instructions.
elevator_operation.webm
This plugin allows you to simulate an elevator system inside the Gazebo environment using ROS 2 topics. The elevator supports floor indexing, precise height targeting, and door open/close commands.
Assuming
<namespace>/ = /lift1:Commands
/lift1/cmd_floor(std_msgs/Int32): floor index (0-based) from<floor_heights>/lift1/cmd_z(std_msgs/Float64): absolute target height in meters/lift1/door_open(std_msgs/Bool):true= open,false= close
Status
/lift1/cabin_z(std_msgs/Float64): measured cabin Z/lift1/door_pos(std_msgs/Float64MultiArray):[right, left]joint positions
Click to expand
# Move by floor index ros2 topic pub /lift1/cmd_floor std_msgs/msg/Int32 "{data: 0}" ros2 topic pub /lift1/cmd_floor std_msgs/msg/Int32 "{data: 1}" ros2 topic pub /lift1/cmd_floor std_msgs/msg/Int32 "{data: 2}" # Move to absolute height (meters) ros2 topic pub /lift1/cmd_z std_msgs/msg/Float64 "{data: 3.0}" # Open / close doors ros2 topic pub /lift1/door_open std_msgs/msg/Bool "{data: true}" ros2 topic pub /lift1/door_open std_msgs/msg/Bool "{data: false}" # Monitor ros2 topic echo /lift1/cabin_z ros2 topic echo /lift1/door_pos
-
Door Control (Optional Feature)
This repository includes a Gazebo Door Plugin and a Behavior-Tree-based Door Controller to control doors in the simulated building via ROS 2.
The environment contains 37 doors in total, with the following types:
- Single hinged doors
- Single sliding doors
- Double hinged doors
- Double sliding doors
All doors share the same topic interface; only the namespace changes (e.g.,
/L1_door1,/L1_door2,/L2_door5, ...).Click to expand
Each door is controlled by a dedicated namespace.
Below is an example for the door with namespace/L1_door1:Commands
/L1_door1/door_open(std_msgs/Bool):true→ open the doorfalse→ close the door
Status
/L1_door1/door_pos: current door joint position(s)- For double doors, this typically represents both leaves (e.g.,
[right, left]).
- For double doors, this typically represents both leaves (e.g.,
# Open / close a door (L1_door1) ros2 topic pub /L1_door1/door_open std_msgs/msg/Bool "{data: true}" ros2 topic pub /L1_door1/door_open std_msgs/msg/Bool "{data: false}" # Monitor door joint position(s) ros2 topic echo /L1_door1/door_pos
To control any other door, simply replace L1_door1 with the corresponding door namespace (e.g.,
/L1_door2,/L1_door_stairs1,/L1_door_toilet,/L2_door3, etc.).
In addition to manual topic-based control, this repository provides a Behavior Tree (BT) node that automatically opens and closes doors along the robot’s planned path.
The BT-based door controller is currently configured for the following 1st-floor doors:
/L1_door_stairs1/L1_door_stairs2/L1_door_toilet
All these doors use the same topic interface as in 5.1 (only the namespace changes).
Run the BT door controller with:
ros2 run door_bt door_bt_runner
Make sure the simulation and navigation stack are running, and that the door plugins for the above namespaces are loaded in the world.
The BT node subscribes to the planned path topic:
/plan
This topic should contain the robot’s planned path (e.g., from a global planner) so that the BT can determine whether the path passes through any controlled doors.
When the planned path passes through one of the configured doors:
The BT automatically opens the door (via <door_namespace>/door_open) before the robot reaches it. After the robot has passed the door, the BT automatically closes it.
This allows the robot to traverse routes that include doors without any manual
ros2 topic pubcommands. -
Stair locomotion (Optional Feature)
Once the Stair Behavior Tree (BT) is running, the robot follows this flow:
-
Wait for a target floor request The BT keeps waiting for a message on
/stairs/floor_request. -
Read the current floor When a target floor is received, the BT estimates the robot’s current floor (e.g., from robot pose / height in the world frame).
-
Decide what to do (up / down / stay)
target_floor > current_floor→ go uptarget_floor < current_floor→ go downtarget_floor == current_floor→ no movement (already on target floor)
-
Navigate to the stair entry
The BT selects the appropriate stair entry pose for the current step and navigates there using Nav2. -
Align and execute stair locomotion
The robot aligns itself with the stairs, performs stair locomotion (mid/top), and runs the landing motions (move/turn) as configured. -
Multi-floor moves (repeat per floor)
If the target floor is more than one level away, the BT repeats the “one-floor stair step” sequence until it reaches the target floor.
ros2 launch quadruped_nav2 quadruped_nav2_bringup.launch.py
- Launches Nav2-related nodes for path planning and navigation.
- Used to move the robot to the stair entry pose before starting stair locomotion.
ros2 run stair_bt stair_bt_runner --ros-args -p bt_xml_file:="$(ros2 pkg prefix stair_bt)/share/stair_bt/bt_trees/stairs.xml"- Runs the stair locomotion Behavior Tree.
- The BT waits for
/stairs/floor_requestand, once received, executes the stair locomotion sequence.
ros2 topic pub /stairs/floor_request std_msgs/msg/Int32 "{data: 1}"- Sends the target floor as a trigger.
- Example above requests floor 1.
-
This example demonstrates how to control the robot's movement using keyboard commands.
-
Activate the Robot: In the terminal where you launched the
rl_action commandnode, enter the following commands to activate the robot and enable navigation mode:(csuite) activation true (csuite) navigation true
-
Launch Keyboard Teleoperation: Open a new terminal and run the
teleop_twist_keyboardnode to control the robot with your keyboard.ros2 run teleop_twist_keyboard teleop_twist_keyboard
You can now move the robot using the capital keys displayed in the terminal (e.g.,
U,I,O,J,K,L,M,<,>) -
Deactivate the Robot: When you are finished, deactivate the robot and exit the command interface by returning to the
rl_actionterminal and entering:(csuite) activation false (csuite) quit
-
Navigation integration
- Add navigation capability using either the
Nav2package or by integrating with unitree_go2_nav
- Add navigation capability using either the
-
OpenRMF support
- Incorporate Open-RMF functionality for fleet and task management
-
RL model update
- Modify reinforcement learning to account for LiDAR weight
This project is licensed under the MIT License. See the LICENSE file for details.
In addition, the following components are based on forked repositories with modifications:
- Controller module is forked and adapted from fan-ziqi/rl_sar
This component follows the license of the original repository.
Maintainer: Sanghyun Kim (kim87@khu.ac.kr)