colcon build crashes because of memory overflow

[TwooSix]

I compiling the project with 16GB of memory plus 16GB of swap, but it still crashes due to the memory and swap being fully occupied...

[fong0802]

Same problem here, I have 8gb ram. Always happen around 4 minutes build time.

[Mechazo11]

@fong0802 for your case, try not to have any chrome, mozilla like applications open. Increase your swap size to 24 Gb and see what happens.

@TwooSix can you post the crash message?

[Mechazo11]

@TwooSix @fong0802 try to build only the ros2_orb_slam3 package using this command

colcon build --packages-select ros2_orb_slam3

[fong0802]

@Mechazo11 I setup 24 Gb more swap memory. I dint notice that i alrdy had 4 Gb swap. So total swap I setup is about 30Gb as shown by system monitor. Before I start to build, I directly switched off GUI.

I build the whole package with:
colcon build --symlink-installl

At the end, the package is sucessfully build with whole lot of warning, too much for me to keep track of.

Then, i try the example provided.
The python node giving out error "no module named natsort"

After I installed the natsort module, it is working now.

If possible can you share briefly how you make it work in ROS2 humble?

I am trying to implement ORB-SLAM on an Intel single board computer that is about 3 times more powerful than Raspberry pi 4.
Since there are my own image processing algorithm running in it, I might need to port a light weight version myself.
I am quite new to ROS. Any advice on that is appreciated. Thanks.

[Mechazo11]

Hi @fong0802

Awesome, I am glad you got to fix the issue.

As I believe you are going to be using GitHub more frequently, consider picking up some of the Markdown syntaxes. This will help to differentiate code from text. An excellent tutorial is here

Now let me try to answer some of your queries

• If possible can you share briefly how you make it work in ROS2 humble?

In the README.md file, I mentioned the ROS 1 port of ORB SLAM3 made by thien94 that inspired ros2_orb_slam3. The basic premise is simple, Dr. Campos`s orb-slam3 framework is essentially an internally shared library of the ros2_orb_slam3 package.

You can see this arrangement in CmakeLists.txt file, more specifically at line 72. Whenever you update any files in ORB SLAM3 say, you wanted to add a file called CommonStructs.h to collect some Struct objects used throughout, you would do the following

add_library(orb_slam3_lib SHARED
  orb_slam3/src/System.cc
  orb_slam3/src/Tracking.cc
  orb_slam3/src/LocalMapping.cc
  orb_slam3/src/LoopClosing.cc
  orb_slam3/src/ORBextractor.cc
  orb_slam3/src/ORBmatcher.cc
  orb_slam3/src/FrameDrawer.cc
  orb_slam3/src/Converter.cc
  orb_slam3/src/MapPoint.cc
  orb_slam3/src/KeyFrame.cc
  orb_slam3/src/Atlas.cc
  orb_slam3/src/Map.cc
  orb_slam3/src/MapDrawer.cc
  orb_slam3/src/Optimizer.cc
  orb_slam3/src/Frame.cc
  orb_slam3/src/KeyFrameDatabase.cc
  orb_slam3/src/Sim3Solver.cc
  orb_slam3/src/Viewer.cc
  orb_slam3/src/ImuTypes.cc
  orb_slam3/src/G2oTypes.cc
  orb_slam3/src/CameraModels/Pinhole.cpp
  orb_slam3/src/CameraModels/KannalaBrandt8.cpp
  orb_slam3/src/OptimizableTypes.cpp
  orb_slam3/src/MLPnPsolver.cpp
  orb_slam3/src/GeometricTools.cc
  orb_slam3/src/TwoViewReconstruction.cc
  orb_slam3/src/Config.cc
  orb_slam3/src/Settings.cc
  orb_slam3/include/System.h
  orb_slam3/include/Tracking.h
  orb_slam3/include/LocalMapping.h
  orb_slam3/include/LoopClosing.h
  orb_slam3/include/ORBextractor.h
  orb_slam3/include/ORBmatcher.h
  orb_slam3/include/FrameDrawer.h
  orb_slam3/include/Converter.h
  orb_slam3/include/MapPoint.h
  orb_slam3/include/KeyFrame.h
  orb_slam3/include/Atlas.h
  orb_slam3/include/Map.h
  orb_slam3/include/MapDrawer.h
  orb_slam3/include/Optimizer.h
  orb_slam3/include/Frame.h
  orb_slam3/include/KeyFrameDatabase.h
  orb_slam3/include/Sim3Solver.h
  orb_slam3/include/Viewer.h
  orb_slam3/include/ImuTypes.h
  orb_slam3/include/G2oTypes.h
  orb_slam3/include/CameraModels/GeometricCamera.h
  orb_slam3/include/CameraModels/Pinhole.h
  orb_slam3/include/CameraModels/KannalaBrandt8.h
  orb_slam3/include/OptimizableTypes.h
  orb_slam3/include/MLPnPsolver.h
  orb_slam3/include/GeometricTools.h
  orb_slam3/include/TwoViewReconstruction.h
  orb_slam3/include/SerializationUtils.h
  orb_slam3/include/Config.h
  orb_slam3/include/Settings.h
  orb_slam3/include/CommonStructs.h
)

The rest is run-of-the mill Cmake rules of adding library, creating executables and linking executables with internal shared and external libraries. Check my CmakeLists.txt, that would be your ideal starting point.

• I am trying to implement ORB-SLAM on an Intel single board computer that is about 3 times more powerful than Raspberry pi 4.*

ORB-SLAM3 is a CPU-intensive library which in my experience, did not run well in Raspberry pi 4 like SoCs. I would strongly recommend going with at least the new Orin Nano board. Then it may be possible to run your image processing algorithms on GPU, use the python node to then transfer those results to ORB SLAM3 leaving most of the CPU computation powers for running ORB SLAM3.

Hope these suggestions helped. Good luck with your work.

With best,
@Mechazo11

Updated 07/15/2024

[fong0802]

@Mechazo11 Thanks for your help!! These give me a good starting point.

Good luck with your publications too.