How to Create cuVGL Maps for a Custom Robot (Using ZED Camera)?

Robotics & Edge Computing Robotics - Isaac Isaac ROS
1

I’m working on a system for map creation and localization using VSLAM, NVBLOX, and cuVGL. While I can load NVBLOX and VSLAM successfully, I’m having trouble running cuVGL for map creation.
I followed these guides:
tutorial_mapping_and_localization, tutorial_map_creation](Tutorial for cuVGL Map Creation — isaac_ros_docs documentation), isaac_ros_visual_global_localization

However, these tutorials seem specific to Nova, whereas I’m trying to use them with my own robot equipped with a ZED camera. After examining the code, I found the isaac_mapping_ros package (which doesn’t seem to be available on GitHub). It appears tightly coupled with the Nova ROS bag implementation, and I can’t provide my own custom ROS bags. While the create_map_offline.py in isaac_mapping_ros package is python script that can be viewed and copied, it runs other complied scripts that are coupled to nova. Does anyone know of alternative tutorials or methods for generating cuVGL maps (and later localizing on them) using ZED cameras? Any insights would be greatly appreciated!

After using the cuVGL maps for localization I want to create 2D occupancy map. I noticed the run_nvblox step in create_map_offline.py in isaac_mapping_ros package, but it suffers from the same issue; it is complied code and nvblox documentation does not write how to create a 2D map in image file (like jpg, png, etc…)

2

Hi @liorst

Right now we don’t have a documentation for the ZED camera, but you can check with rqt_graph if all nodes are properly connected.

If not, you can may make a new own launch file remapping all topics.

Let me know,
Raffaello

3

To create cuVGL maps with a custom robot using a ZED camera, first integrate the ZED ROS wrapper for camera data and use VSLAM (e.g., ORB-SLAM2 or ZED VINS) for pose estimation and map building. Modify the isaac_mapping_ros package to work with your setup, especially the create_map_offline.py script, to adapt it for your custom robot and ZED camera data. For generating 2D occupancy maps, downsample the 3D data using tools like costmap_2d or GMapping in ROS. If compiled code limits functionality, explore alternatives or adjust the existing code to process the data for cuVGL map creation and localization.