ORB-SLAM3 Extension: Real-Time Sparse Map to Occupancy Grid (.yaml/.pgm) for ROS 2 Navigation

Projects
, ,
1

Hi :waving_hand:,

I’m Harishma Prakash, an undergraduate from Chennai Institute of Technology, India.

I’ve extended ORB-SLAM3 to include a real-time occupancy grid generator. It takes the sparse 3D map and outputs a ROS 2-compatible .pgm and map.yaml automatically during SLAM execution — no extra commands needed.

GitHub Repository:

Key Features:

  • Runs as a background thread inside ORB-SLAM3
  • Generates .png, .pgm, and map.yaml during SLAM
  • Uses voxel filtering + 2D convex hull + raycasting
  • Generates ROS2 compatible maps

It’s still basic but functional. I would love to hear feedback or suggestions.
Thanks!

— Harishma Prakash

1 Like
ROS News for the Week of July 21st, 2025
Open Robotics News for the Week of July 13th, 2025
2

This looks neat! Do you have any videos or example maps that this creates? That might help folks get interested and involved in your work!

1 Like
3

Thank you! I have attached an example of the working with kitti dataset.

orbslam3_map

kittiexample

The below image is the attempt of creating occupancy grid using the generally used method of directly flattening the sparse map points to the 2D plane:

direct_sparse_to_occupancy
direct_sparse_to_occupancy1000×1000 104 KB

4

Hi, I’m having difficulty setting up ORB-SLAM3. Would you mind sharing the steps you followed to install and configure it, including the prerequisite dependencies and Ubuntu version? I’m trying to run it on a VirtualBox VM with a monocular camera, and still stuck and confused with it.

5

Hi! I followed the setup instruction in their original github,

but in Ubuntu 20 and used openCV 4.3 instead 4.4. Since I too couldn’t setup ORBSLAM3 properly with version 4.4, i moved to 4.3 which is more compatible with Ubuntu 20 in a docker container. Rest of the pre-requisites were implemented as per their official documentation in github.