A TurtleBot3 Burger drives itself through the multi-room TurtleBot3 House world in Gazebo Classic 11, using the ROS 2 Humble Nav2 stack. A small Python script sends it to three goals in different rooms, and Nav2 plans the paths and steers around the walls and furniture on the way.
Screen recording (Gazebo + RViz + terminal): Watch on Google Drive
Robot moving through the house floorplan across the three goals.
- Loads the TurtleBot3 House world and spawns a TurtleBot3 Burger (2D LiDAR + wheel odometry).
- Localizes on a pre-built map with AMCL, starting from a known pose of
(-2.0, 1.0, 0°). - Plans global paths with
NavfnPlanner(A* enabled) and follows them with the DWB local planner, which avoids obstacles using the local costmap. - Runs a three-goal mission through
nav2_simple_commander.BasicNavigatorand prints distance-to-goal and time per goal. - Opens RViz2 with a saved config (map, scan, costmaps, plans, particle cloud, TF).
Scope note: the obstacles here are the static ones in the house (walls, furniture, doorways). I did not add moving or unmapped obstacles, so replanning around something unexpected is not demonstrated in this repo.
| Component | Choice |
|---|---|
| OS | Ubuntu 22.04 |
| ROS | ROS 2 Humble |
| Simulator | Gazebo Classic 11.10.2 |
| Navigation | Nav2 (Humble) |
| Localization | nav2_amcl, differential motion model |
| Global planner | nav2_navfn_planner/NavfnPlanner (A*) |
| Local planner | dwb_core::DWBLocalPlanner |
| Robot | TurtleBot3 Burger |
| Language | Python 3.10 |
TurtleBot3/
├── README.md
├── package.xml # ROS 2 package manifest (ament_python)
├── setup.py / setup.cfg
├── requirements.txt
├── config/
│ ├── nav2_params.yaml # Nav2 config (costmaps, AMCL, planners, controller)
│ └── rviz_config.rviz # RViz2 layout
├── docs/
│ └── troubleshooting.md # Problems I hit and their root causes
├── launch/
│ ├── bringup.launch.py # Gazebo + Nav2 + RViz2
│ ├── navigation.launch.py# Nav2 + RViz2
│ └── simulation.launch.py# Gazebo + robot_state_publisher + spawn
├── maps/
│ ├── map.pgm # Occupancy grid of the house (16 m x 12 m)
│ └── map.yaml # 0.05 m/cell, origin [-8.0, -6.0]
├── scripts/
│ └── verify_setup.sh # Pre-flight dependency check
├── src/
│ ├── navigation_demo.py # Three-goal mission client
│ └── start_robot.py # Launcher that starts Gazebo and Nav2 and cleans up after
└── worlds/
└── turtlebot3_house.world
- Ubuntu 22.04 (x86_64), ROS 2 Humble Desktop at
/opt/ros/humble, Gazebo Classic 11.
sudo apt update && sudo apt install -y \
ros-humble-turtlebot3-gazebo \
ros-humble-turtlebot3-navigation2 \
ros-humble-nav2-bringup \
ros-humble-nav2-simple-commander \
ros-humble-gazebo-ros \
ros-humble-robot-state-publisher \
ros-humble-rviz2 \
python3-transforms3dThe world, map, Nav2 params and RViz config are all in this repo, so nothing else needs downloading.
cd <workspace_directory>
git clone <repository_url>
cd <repository_directory>
source /opt/ros/humble/setup.bash
export TURTLEBOT3_MODEL=burger
export GAZEBO_MODEL_DATABASE_URI="" # stops Gazebo hanging on the dead online model DB
chmod +x scripts/verify_setup.sh src/start_robot.py src/navigation_demo.py launch/*.py
./scripts/verify_setup.shstart_robot.py and simulation.launch.py set GAZEBO_MODEL_DATABASE_URI themselves, so the export is only needed if you launch Gazebo by hand.
You need two terminals.
Terminal 1: simulation + Nav2
cd <path_to_repository>
source /opt/ros/humble/setup.bash
export TURTLEBOT3_MODEL=burger
./src/start_robot.py --with-navThis checks dependencies, clears leftover Gazebo/Nav2 processes and DDS shared-memory files, launches Gazebo with the house world, spawns the robot at (-2.0, 1.0), and starts Nav2 and RViz2. Wait 10-15 s for:
[READY] System is running. Press Ctrl+C to stop simulation safely.
Don't start
start_robot.pytwice. Two simulations collide on Fast-DDS shared-memory ports and Nav2 never comes up.Heads up: the cleanup step runs
pkill -9ongzserver,gzclientandcomponent_container, so it will also kill any other Gazebo or Nav2 you have running. If you prefer, launch withros2 launch launch/bringup.launch.pyinstead.
Terminal 2: mission
cd <path_to_repository>
source /opt/ros/humble/setup.bash
export TURTLEBOT3_MODEL=burger
python3 src/navigation_demo.pyIt waits for /clock and the Nav2 action servers, sets the AMCL initial pose, then sends the three goals in order. The script exits when it's done. Press Ctrl+C in Terminal 1 to shut everything down.
The room names below are just my own labels for places in the house.
| # | Label | X (m) | Y (m) | Yaw | Route |
|---|---|---|---|---|---|
| 1 | North-East room | 4.50 | 2.50 | 0° | Out of the living room through the doorway |
| 2 | South-East room | 6.00 | -1.00 | -90° | Through the connecting hallway |
| 3 | Start / living room | -2.00 | 1.00 | 180° | Back across the whole house |
Each goal has a 90-120 s timeout. The script reports whether Nav2 returned SUCCEEDED.
The saved config (config/rviz_config.rviz) shows: robot model, TF tree (map -> odom -> base_footprint -> base_link -> base_scan), static map, laser scan, AMCL particle cloud, global and local costmaps, the global plan from the planner, and the local trajectory from DWB.
From one of my runs (times depend on your machine's simulation speed):
[3/4] Executing autonomous navigation sequence (3 goals)...
--- Navigating to Waypoint 1 (North-East Master Room): x=4.50, y=2.50 ---
[SUCCESS] Reached Waypoint 1 (North-East Master Room) in 39.1s!
--- Navigating to Waypoint 2 (South-East Sunroom): x=6.00, y=-1.00 ---
[SUCCESS] Reached Waypoint 2 (South-East Sunroom) in 18.7s!
--- Navigating to Waypoint 3 (Return to Living Room Base): x=-2.00, y=1.00 ---
[SUCCESS] Reached Waypoint 3 (Return to Living Room Base) in 48.1s!
- Waypoint 1 : [PASS] (39.1s)
- Waypoint 2 : [PASS] (18.7s)
- Waypoint 3 : [PASS] (48.1s)
[DEMO COMPLETE] All navigation goals succeeded!
Full write-ups are in docs/troubleshooting.md. The main ones:
- Gazebo hangs or exits with code 255. Gazebo Classic tries to reach
models.gazebosim.org, which no longer serves models. Fix:export GAZEBO_MODEL_DATABASE_URI="". - Demo stuck on "Waiting for Nav2 stack lifecycle nodes to activate". A leftover Gazebo/Nav2 process was holding Fast-DDS shared-memory ports. Fix:
This is what
pkill -9 -f "gzserver|gzclient|component_container" || true rm -f /dev/shm/fastrtps_* /dev/shm/sem.fastrtps_*
start_robot.pyautomates. - TF extrapolation errors. Some node was using wall-clock time instead of Gazebo's. Fix:
use_sim_time:=trueeverywhere, includingnav2_params.yaml.
Useful checks when something looks off:
ros2 topic echo /clock --once
ros2 node list
ros2 lifecycle get /amcl
ros2 lifecycle get /bt_navigator
ros2 run tf2_ros tf2_echo odom base_link- Simulation only. Real hardware would need motor drivers, LiDAR calibration and slip handling.
- Only static obstacles were tested. I haven't tried moving or newly placed obstacles, so dynamic replanning is untested.
- Needs a pre-built map. Unknown environments would need SLAM (
slam_toolboxor Cartographer). - Classical planning only (Navfn + DWB), with the default behavior tree and recovery behaviors.
Apache 2.0. See LICENSE.

