Ongoing development.

ROS wrapper for ORB-SLAM3

A ROS wrapper for ORB-SLAM3. The main idea is to use the ORB-SLAM3 as a standalone library and interface with it instead of putting everything in one package.

• Pros:
  • Easy to update the ORB-SLAM3 library (currently in V0.3 Beta).
  • Easy to plug in different variants (and there are many) that are not built for ROS (hopefully).
• Cons:
  • Might be more difficult to spot bugs.
  • Development involves more steps (1. Make changes in ORB-SLAM3 library -> 2. Build ORB-SLAM3 -> 3. Change the ROS-wrapper if necessary -> 4. Test).
  • Might break when dependencies or upstream changes.

To-do

• Add nodes for each supported sensor type (mono, mono + imu, stereo, stereo + imu, rgbd).
• Add launch files for the nodes.
• Add ROS publishers for pose, pointcloud, path.
• Add ROS publisher for tracking image (require changes in ORB-SLAM3 to be merged).
• Test with different variants of ORB-SLAM2/3, such as DynaSLAM or SuperPoint-SLAM.

Current known limitations

• For mono+imu and stereo+imu cases, the world frame seems to change arbitrarily each time a big Bundle Adjustment is invoked. As a result, you will see the pose data suddenly jumps around, but actually it is the world frame that teleports. The issue is raised here.
• Currently there is no public API to obtain the tracking image from ORB-SLAM3, so wait for this PR to be merged.
• Currently there is no public API to obtain the full map points and other data to debug.

Installation

First install ORB-SLAM3 normally with all of its dependencies (any location is fine) then install this package in a catkin build environment.

1. ORB-SLAM3 (original or other variants)

• Install the prerequisites.
• Build and install ORB-SLAM3. Any location is fine (default directory that I use later on is the home folder ~):

cd ~
git clone https://github.com/UZ-SLAMLab/ORB_SLAM3.git ORB_SLAM3
cd ORB_SLAM3
chmod +x build.sh
./build.sh

• Make sure that libORB_SLAM3.so is created in the ORB_SLAM3/lib folder. If not, check the issue list from the original repo and rebuild the package.

2. orb_slam3_ros_wrapper (this package)

• Clone the package. Note that it should be a catkin build workspace.

cd ~/catkin_ws/src/ # Or the name of your workspace
git clone https://github.com/thien94/orb_slam3_ros_wrapper.git

• Open CMakeLists.txt and change the directory that leads to ORB-SLAM3 library at the beginning of the file (default is home folder)

cd ~/catkin_ws/src/orb_slam3_ros_wrapper/
nano CMakeLists.txt

# Change this to your installation of ORB-SLAM3. Default is ~/
set(ORB_SLAM3_DIR
   $ENV{HOME}/ORB_SLAM3
)

• Build the package normally.

cd ~/catkin_ws/
catkin build

• Unzip the ORBvoc.txt file in the config folder in this package. Alternatively, you can change the voc_file param in the launch file to point to the right folder.

cd ~/catkin_ws/src/orb_slam3_ros_wrapper/config
tar -xf ORBvoc.txt.tar.gz

• Install hector-trajectory-server to visualize the trajectory.

sudo apt install ros-[kinetic/melodic]-hector-trajectory-server

• If everything works fine, you can now try the different launch files in the launch folder.

3. How to run

Example with EuRoC dataset:

# In one terminal
roslaunch orb_slam3_ros_wrapper orb_slam3_mono_inertial_euroc.launch
# In another terminal
rosbag play MH_01_easy.bag

Similarly for other sensor types.

Topics

The following topics are published by each node:

• /orb_slam3_ros/map_points (PointCloud2) containing all keypoints being tracked.
• /orb_slam3_ros/camera (PoseStamped): current pose of the camera in the world frame, as returned by ORB-SLAM3 with the world coordinate transformed to conform the ROS standard.
• tf: transformation from the camera fraame to the world frame.