RTAB-Map Documentation

The imufilter module provides two primary filters for fusing IMU data (angular velocities, accelerations, and optionally magnetic readings) into an orientation:

• MadgwickFilter: Fuses sensor readings into an orientation. Note: This filter is licensed under GPL.
• ComplementaryFilter: Fuses sensor readings into an orientation quaternion using a complementary fusion approach. This filter is licensed under BSD.

The MadgwickFilter is licensed under GPL. If you need to avoid GPL licensing in your project, you can disable this filter during the compilation process by passing the WITH_MADGWICK flag to CMake.

cmake -DWITH_MADGWICK=OFF ..

Use the provided shell scripts to run the SLAM processing. Results are written to results/euroc, results/kitti, and results/tum subfolders.

Note: Running all datasets with all odometry approaches can take over 10 hours. You can modify run_all.sh to limit the scope.

• Process all datasets: ./run_all.sh
• Process KITTI sequence 07 with F2M odometry: ./run_kitti_datasets.sh f2m 0 0 07

Before processing TUM datasets, you must synchronize the RGB and depth images using the associate.py script. This creates rgb_sync and depth_sync folders.

Usage: python associate.py rgb.txt depth.txt

The pose_graph_2d executable solves 2D pose graph optimization problems using Ceres. It accepts problem definitions in the g2o format.

When run, the executable prints the Ceres solver summary to the console and generates two output files:

• poses_original.txt: The initial poses before optimization.
• poses_optimized.txt: The optimized poses.

Each output file contains poses in the following format, sorted by pose_id in ascending order:

pose_id x y yaw_radians

/path/to/bin/pose_graph_2d --input /path/to/dataset/dataset.g2o

To use RTAB-Map within the Robot Operating System (ROS), you can install the corresponding ROS binaries. The package name follows the pattern ros-$ROS_DISTRO-rtabmap.

Available ROS distributions include:

• ROS 1: Noetic
• ROS 2: Humble, Jazzy, Kilted, Lyrical, and Rolling

For detailed installation instructions and examples, refer to the RTAB-Map wiki. For ROS-specific integration, visit the ROS wiki page.

The pose_graph_3d executable performs 3D pose graph optimization using Ceres. It accepts problem definitions in the g2o format that use quaternions for orientation.

Upon completion, the tool prints a Ceres solver summary and generates two files:

• poses_original.txt: The initial poses.
• poses_optimized.txt: The optimized poses.

Both files follow this format (sorted by pose_id in ascending order):

pose_id x y z q_x q_y q_z q_w

/path/to/bin/pose_graph_3d --input /path/to/dataset/dataset.g2o

This Docker image is used to reproduce the results presented in the paper "RTAB-Map as an Open-Source Lidar and Visual SLAM Library for Large-Scale and Long-Term Online Operation" (Labbé and Michaud, 2018) for KITTI, EuRoC, and TUM datasets.

You can either pull the pre-built image or build it locally from the provided Dockerfile.

# Pull the pre-built image
docker pull introlab3it/rtabmap:jfr2018

# Or build it locally
cd rtabmap/docker/jfr2018
docker build -t introlab3it/rtabmap:jfr2018 .

To use the reproduction scripts, you must organize your datasets in a datasets subfolder relative to the scripts. The directory structure must follow this pattern:

datasets/
  kitti/
    devkit
    dataset/sequences/00
    ...
  euroc/
    MH_01_easy
    ...
  tum/
    rgbd_dataset_freiburg1_desk
    ...

run_all.sh
run_kitti_datasets.sh
run_euroc_datasets.sh
run_tum_datasets.sh

You can visualize the difference between initial and optimized poses using the provided Python script. Use the --axes_equal flag if you want to enable equal axes in the visualization.

/path/to/repo/examples/slam/pose_graph_3d/plot_results.py --optimized_poses ./poses_optimized.txt --initial_poses ./poses_original.txt