Best practice for validation and testing?

[rakuwaku]

Hi,

  I have completed simulation testing with sim sensors using Rtabmap. The next step is validation on real hardware with real cameras. Is there a best practice in the verification and tuning of rtabmap on hardware?

  I have SLAM running and everything runs OK on startup. I want to limit tests using hardware as much as possible so I'd like to gather all the real data collected and be able to rerun them in sim for fine tuning.

I'm using ROS. Does this mean I should just bag all the camera, and odometry sources? I would also have a copy of the rtabmap.db thats generated.

Is there anything else I'm missing?

[matlabbe]

Recording a bag with all data input to rtabmap is a good idea to debug afterwards and test different parameters by replaying the rosbag. In general, we don't start rtabmap node when recording to keep TF clean of the mapping process (otherwise when replaying the bag with mapping, you may get two /map -> /odom frames that would cause strange glitch in mapping).

If odometry is debugged, tuned and work correctly, you can use the generated database as input in rtabmap standalone to just tune the mapping part (loop closure detection, grid generation...).

cheers,
Mathieu