rtabmap with simulated topics

[Parastou Bakhtiari]

https://drive.google.com/drive/folders/15fMnaFpHj2YdzMMj__Vn3BsuJa1bxfLN?usp=sharing

http://https://drive.google.com/drive/folders/15fMnaFpHj2YdzMMj__Vn3BsuJa1bxfLN?usp=sharing

Hi,
I hope you are doing well.

I’m running RTAB-Map (ROS 2 Humble) using simulated RGB-D topics from MAVS (Mississippi State University’s agricultural simulation).
RTAB-Map launches and subscribes successfully, but the map is not consistent and I receive repeated warnings about time-sync and invalid depth values.

in the attached link you can find the shared ros2bag file recorded from the running ros2 node in simulation.
this is the command I used:
ros2 launch rtabmap_launch rtabmap.launch.py    args:="--delete_db_on_start"    frame_id:=zed_left_camera_optical_frame    odom_topic:=/mavs/odometry_true    visual_odometry:=false    rgb_topic:=/mavs/camera    depth_topic:=/mavs/depth/depth_registered    camera_info_topic:=/mavs/camera_info    wait_imu_to_init:=true    imu_topic:=/mavs/imu    rviz:=true    approx_sync:=true    rgbd_sync:=true    approx_rgbd_sync:=true    topic_queue_size:=10    use_sim_time:=true

(Note: without launching my urdf)

the warnings are:
[rgbd_sync] The time difference between rgb and depth frames is high (diff≈0.43s)
[rtabmap] Cloud with only NaN values created!
[rtabmap_viz] Received odom topic with child_frame_id not set! Using "zed_left_camera_optical_frame" as base frame.
[rtabmap] (getting transform zed_left_camera_optical_frame -> imu_link) ... does not exist.


Would you please help me to solve the issue?

[matlabbe]

Hi,

There is an issue with the camera topics generated. The /mavs/camera and /mavs/depth/depth_registered topics have exactly the same stamp, but not /mavs/camera_info. You should fix /mavs/camera_info so that it publishes with the same stamp than the other topics, then use approx_rgbd_sync:=false too.

For the odometry topic, there is a missing child_frame_id.

The depth looks really weird, it is very difficult to know if depth is fine or not. Could you use a simulated world that is not random points? Like https://docs.px4.io/v1.12/en/simulation/gazebo_worlds.html