I was wondering if it would be possible to do something like this: Vision-based Autonomous Navigation and Mapping with a Small Quadrotor without using lidar/lasers.

I have tried ORB-SLAM, but the maps that are generated are too sparse for a person to get any meaningful sense from the map.

  • $\begingroup$ here is the answer: youtube.com/watch?v=yfKccCmmMsM $\endgroup$ Nov 19, 2019 at 3:41
  • 1
    $\begingroup$ The video you linked to is only using Cameras and IMU. If you want a more dense map then you need to combine ORB-SLAM with a dense mapping algorithm like Octomap. $\endgroup$
    – edwinem
    Nov 19, 2019 at 18:55

1 Answer 1


Structure SLAM paper by Zhou et al. (2022) maybe a good starting point for your project. The paper showcases dense mapping with 3d cuboid detection.

YouTube link: https://www.youtube.com/watch?v=nBbGTFeUh88

GitHub code: https://github.com/benchun123/point-plane-object-SLAM

From my initial read, Structure SLAM is based upon ORB SLAM2 and extends the ideas of CubeSLAM (the latter is my opinion, not concretely mentioned in the paper). Requires RGB and depth map. Since you have IMU data, you maybe able to combine all three modalities but I am not aware if ORB SLAM2 does this natively.

I would suggest you to start with ORB SLAM3 since it is more accurate than ORB SLAM2. I have released a repository that natively implements ORB SLAM3 in ROS2 Humble


Hope this helps.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.