System Recommendations

[MidKnight]

Hello @matlabbe, greetings!

I am working on a project with an aim of enabling autonomous navigation on a drone using rtabmap_ros with RealSense D435 camera, in RGBD mode.

My plan is to use Seeed Studio Jetson Nano (4GB RAM) as the onboard computer to record ros2 bag data and offload that bag on to my workstation for constructing the 3D map and the 2D OccupanyGrid map needed for nav2. Then feed the constructed map back to the Jetson, running RTABMap in localisation mode with the fed map to perform autonomous navigation using nav2 for planning, mavros for interfacing with PX4, and a PX4 enabled FCU for drone control.

Does this system sound correct to you, is this the best way to achieve what I want? I request you to please provide your opinion and suggestions on this. Any help is greatly appreciated, thanks in advance!

Sensor: Intel RealSense D435 RGBD Camera 

Workstation:
On board Computer:  Seeed studio Jetson Nano

Regards,
MidKnight.

[matlabbe]

You can see my other answer about how to setup the D435 for best results: http://official-rtab-map-forum.206.s1.nabble.com/Improving-RTABMap-performance-tp11035p11040.html

For the computation, I am concerned about the Nano power for fast enough visual odometry when you are going to trying localization mode. You may consider looking at Isaac VO/VIO approach that can be more efficiently on Nano:
https://github.com/introlab/rtabmap_ros/blob/f5a3e487056e399b834971ef6a64126843666f7f/rtabmap_demos/launch/isaac/isaac_sim_vslam_demo.launch.py#L232C8-L261
or other CPU efficient approaches like msckf_vio.

cheers,
Mathieu