Skip to main content
ORB-SLAM3 provides ROS integration for processing live camera and IMU data. This guide covers building ROS nodes, configuration, and running with ROS bags or live sensors.

​
Overview

ORB-SLAM3 ROS nodes support all sensor configurations:
  • Mono: Monocular camera
  • Mono_Inertial: Monocular camera with IMU
  • Stereo: Stereo camera pair
  • Stereo_Inertial: Stereo cameras with IMU
  • RGBD: RGB-D depth camera
  • MonoAR: Monocular with augmented reality demo
ROS integration has been tested with ROS Melodic on Ubuntu 18.04. It should work with other ROS distributions with minimal modifications.

​
Building ROS Nodes

​
Prerequisites

Before building ROS nodes:
  1. Install ROS: Follow instructions at ros.org
  2. Build ORB-SLAM3: Complete the main build process first
  3. Source ROS: source /opt/ros/melodic/setup.bash

​
Setup ROS Package Path

1

Add ORB-SLAM3 to ROS package path

Edit your ~/.bashrc file:
gedit ~/.bashrc
2

Append the path

Add this line at the end (replace PATH with your ORB-SLAM3 directory):
export ROS_PACKAGE_PATH=${ROS_PACKAGE_PATH}:PATH/ORB_SLAM3/Examples_old/ROS
3

Source the updated bashrc

source ~/.bashrc

​
Build the ROS Nodes

Use the provided build script: 
cd ORB_SLAM3
chmod +x build_ros.sh
./build_ros.sh
This script:
  • Navigates to Examples_old/ROS/ORB_SLAM3
  • Creates a build directory
  • Runs CMake with ROS_BUILD_TYPE=Release
  • Compiles all ROS nodes

​
Build Output

After successful compilation, executables are created: 
Examples_old/ROS/ORB_SLAM3/build/
├── Mono                # Monocular node
├── MonoAR              # Monocular AR demo
├── Stereo              # Stereo node
├── RGBD                # RGB-D node
├── Mono_Inertial       # Monocular-Inertial node
└── Stereo_Inertial     # Stereo-Inertial node

​
ROS Node Types

​
Monocular Node

Processes monocular camera images from /camera/image_raw: 
rosrun ORB_SLAM3 Mono \
  PATH_TO_VOCABULARY \
  PATH_TO_SETTINGS_FILE
Example: 
rosrun ORB_SLAM3 Mono \
  Vocabulary/ORBvoc.txt \
  Examples/Monocular/EuRoC.yaml
Topics subscribed:
  • /camera/image_raw (sensor_msgs/Image)

​
Monocular-Inertial Node

Processes monocular images with IMU data: 
rosrun ORB_SLAM3 Mono_Inertial \
  PATH_TO_VOCABULARY \
  PATH_TO_SETTINGS_FILE \
  [EQUALIZATION]
Example: 
rosrun ORB_SLAM3 Mono_Inertial \
  Vocabulary/ORBvoc.txt \
  Examples/Monocular-Inertial/EuRoC.yaml
Topics subscribed:
  • /camera/image_raw (sensor_msgs/Image)
  • /imu (sensor_msgs/Imu)
Optional argument:
  • EQUALIZATION: Set to true to apply CLAHE histogram equalization (useful for TUM-VI dataset)
For challenging lighting conditions or low-contrast scenes, enable equalization:
rosrun ORB_SLAM3 Mono_Inertial Vocabulary/ORBvoc.txt config.yaml true

​
Stereo Node

Processes stereo camera pair: 
rosrun ORB_SLAM3 Stereo \
  PATH_TO_VOCABULARY \
  PATH_TO_SETTINGS_FILE \
  ONLINE_RECTIFICATION
Example: 
rosrun ORB_SLAM3 Stereo \
  Vocabulary/ORBvoc.txt \
  Examples/Stereo/EuRoC.yaml \
  true
Topics subscribed:
  • /camera/left/image_raw (sensor_msgs/Image)
  • /camera/right/image_raw (sensor_msgs/Image)
Online rectification:
  • true: Node rectifies images (provide rectification matrices in YAML)
  • false: Images must be pre-rectified
  • Not needed for fisheye cameras (works with original images)

​
Stereo-Inertial Node

Processes stereo cameras with IMU: 
rosrun ORB_SLAM3 Stereo_Inertial \
  PATH_TO_VOCABULARY \
  PATH_TO_SETTINGS_FILE \
  ONLINE_RECTIFICATION \
  [EQUALIZATION]
Example: 
rosrun ORB_SLAM3 Stereo_Inertial \
  Vocabulary/ORBvoc.txt \
  Examples/Stereo-Inertial/EuRoC.yaml \
  true
Topics subscribed:
  • /camera/left/image_raw (sensor_msgs/Image)
  • /camera/right/image_raw (sensor_msgs/Image)
  • /imu (sensor_msgs/Imu)

​
RGB-D Node

Processes RGB-D depth camera: 
rosrun ORB_SLAM3 RGBD \
  PATH_TO_VOCABULARY \
  PATH_TO_SETTINGS_FILE
Example: 
rosrun ORB_SLAM3 RGBD \
  Vocabulary/ORBvoc.txt \
  Examples/RGB-D/TUM1.yaml
Topics subscribed:
  • /camera/rgb/image_raw (sensor_msgs/Image)
  • /camera/depth_registered/image_raw (sensor_msgs/Image)
Depth images must be registered (aligned) to the RGB frame. Configure RGBD.DepthMapFactor correctly in your YAML file.

​
Topic Configuration

By default, ROS nodes subscribe to standard topics. Remap topics when running with different camera drivers:

​
Topic Remapping

# Remap monocular topic
rosrun ORB_SLAM3 Mono Vocabulary/ORBvoc.txt config.yaml \
  /camera/image_raw:=/usb_cam/image_raw

# Remap stereo topics
rosrun ORB_SLAM3 Stereo Vocabulary/ORBvoc.txt config.yaml true \
  /camera/left/image_raw:=/stereo/left/image_rect \
  /camera/right/image_raw:=/stereo/right/image_rect

# Remap IMU topic
rosrun ORB_SLAM3 Mono_Inertial Vocabulary/ORBvoc.txt config.yaml \
  /imu:=/mavros/imu/data

​
Default Topic Names

Monocular

  • /camera/image_raw

Stereo

  • /camera/left/image_raw
  • /camera/right/image_raw

RGB-D

  • /camera/rgb/image_raw
  • /camera/depth_registered/image_raw

IMU

  • /imu

​
Running with ROS Bags

Test ORB-SLAM3 with recorded ROS bags:

​
Complete Example: EuRoC Stereo-Inertial

1

Download EuRoC dataset

wget http://robotics.ethz.ch/~asl-datasets/ijrr_euroc_mav_dataset/\
machine_hall/MH_01_easy/MH_01_easy.bag
2

Start roscore

In terminal 1:
roscore
3

Launch ORB-SLAM3 node

In terminal 2:
rosrun ORB_SLAM3 Stereo_Inertial \
  Vocabulary/ORBvoc.txt \
  Examples/Stereo-Inertial/EuRoC.yaml \
  true
4

Play the rosbag

In terminal 3, wait for ORB-SLAM3 to load vocabulary, then:
rosbag play --pause MH_01_easy.bag \
  /cam0/image_raw:=/camera/left/image_raw \
  /cam1/image_raw:=/camera/right/image_raw \
  /imu0:=/imu
Press space to start playback.

​
Playback Tips

# Play at normal speed
rosbag play dataset.bag

​
TUM-VI Dataset Rebag Issue

TUM-VI rosbags may have playback issues due to large chunk size: 
# Rebag with default chunk size
rosrun rosbag fastrebag.py \
  dataset-room1_512_16.bag \
  dataset-room1_512_16_small_chunks.bag
Then play the rebbagged file normally.

​
Running with Live Cameras

​
Using RealSense Cameras

For Intel RealSense cameras, use the realsense-ros package: 
# Install RealSense ROS wrapper
sudo apt-get install ros-melodic-realsense2-camera

# Launch RealSense node (D435i example)
roslaunch realsense2_camera rs_camera.launch \
  enable_gyro:=true \
  enable_accel:=true \
  unite_imu_method:=linear_interpolation

# In another terminal, run ORB-SLAM3
rosrun ORB_SLAM3 Stereo_Inertial \
  Vocabulary/ORBvoc.txt \
  Examples/Stereo-Inertial/RealSense_D435i.yaml \
  false \
  /camera/infra1/image_rect_raw:=/camera/left/image_raw \
  /camera/infra2/image_rect_raw:=/camera/right/image_raw \
  /camera/imu:=/imu

​
USB Webcam

For simple USB cameras: 
# Install usb_cam package
sudo apt-get install ros-melodic-usb-cam

# Launch camera node
rosrun usb_cam usb_cam_node

# Run ORB-SLAM3 monocular
rosrun ORB_SLAM3 Mono \
  Vocabulary/ORBvoc.txt \
  my_webcam_config.yaml \
  /usb_cam/image_raw:=/camera/image_raw

​
Troubleshooting

Cause: ROS can’t find ORB-SLAM3 package.Solution:
# Check ROS_PACKAGE_PATH
echo $ROS_PACKAGE_PATH

# Should include: /path/to/ORB_SLAM3/Examples_old/ROS

# If not, add to ~/.bashrc
export ROS_PACKAGE_PATH=${ROS_PACKAGE_PATH}:/path/to/ORB_SLAM3/Examples_old/ROS
source ~/.bashrc
Cause: Text vocabulary file is slow to parse.Solution: Be patient (30-60 seconds). Consider converting to binary format for faster loading.
# Wait for this message before playing rosbag
# "ORB-SLAM3 System is ready."
Cause: Topic mismatch or message type incompatibility.Solution:
# List active topics
rostopic list

# Check message type
rostopic info /camera/image_raw

# Should be: sensor_msgs/Image

# Monitor topic
rostopic hz /camera/image_raw

# Use correct topic remapping
rosrun ORB_SLAM3 Mono vocab.txt config.yaml \
  /camera/image_raw:=/your/actual/topic
Cause: Insufficient IMU data, poor calibration, or timestamp issues.Solution:
  • Ensure IMU publishing at correct frequency (check config: IMU.Frequency)
  • Verify camera-IMU timestamps are synchronized
  • Check IMU calibration parameters in YAML
  • Provide sufficient motion during initialization (15-20 seconds)
# Check IMU topic
rostopic hz /imu

# Should match IMU.Frequency in config (e.g., 200 Hz)
Cause: Incorrect camera calibration or configuration.Solution:
  • Verify camera calibration parameters in YAML
  • Check image resolution matches config (Camera.width, Camera.height)
  • Ensure vocabulary file loaded correctly
  • Test with lower ORB threshold (ORBextractor.minThFAST: 5)
  • Verify sufficient visual texture in scene
Cause: Stereo synchronization or rectification issues.Solution:
  • Ensure stereo images are properly synchronized
  • Check Stereo.T_c1_c2 transformation is correct
  • Verify baseline (Stereo.ThDepth) is appropriate
  • For pinhole: Enable online rectification or pre-rectify images
# Check stereo synchronization
rosrun image_view stereo_view \
  stereo:=/camera image:=image_raw

​
Performance Optimization

​
Real-Time Performance Tips

1

Reduce ORB features

Lower ORBextractor.nFeatures to 800-1000 for faster processing:
ORBextractor.nFeatures: 800
2

Disable viewer

Run without visualization for maximum performance:
// Modify ROS node or build with viewer disabled
System SLAM(argv[1], argv[2], System::STEREO, false);
3

CPU governor

Set CPU to performance mode:
sudo cpufreq-set -g performance
4

Message throttling

Reduce camera framerate if needed:
rosrun topic_tools throttle messages /camera/image_raw 15.0

​
Saving Trajectory

After running ORB-SLAM3, save the estimated trajectory: 
# In the ORB-SLAM3 terminal, after shutdown
# Trajectories are automatically saved to:
ls -l KeyFrameTrajectory.txt
ls -l CameraTrajectory.txt
Use these files for evaluation against ground truth.

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us