RTAB-Map setup failed on Nvidia Jetson Nano Developer Kit

[Benirus]

Hi everyone!

We are trying to get the T265 and D436 working together on Nvidia Jetson nano dev kit.

Running Ubuntu 18.04, we got realsense2 and RTAB-Map for ROS.

The main issue is the error's produced when trying to listen in on the camera topics (from roslaunch realsense2_camera rs_d435_and_t265.launch align_depth:=true), and the warnings given in console after running roslaunch rtabmap_ros omg.launch depth_registration:=true

First screenshot shows the error's and the second one show the RTAB-Map Gui.

Have we overlooked something major here? Any of this look familiar? A bit worried we lost some dependencies or completely screwed the pooch with the launch file.

[matlabbe]

Hi,

the resolution of the first image is too small, I cannot read the error messages.

For the 3D Map view rendering problem, it is a known problem on latest Jetpack. A solution has been shown here: https://github.com/introlab/rtabmap/issues/427#issuecomment-516914817

However, as explained in that post, if you are under ROS, use RVIZ instead of rtabmapviz... you will save a lot of time and pain from recompiling everything ;)

cheers,
Mathieu

[Benirus]

Hey Mathieu! Thanks for the advice, good to know we missed something fundamentally broken, was going quite mad working on the fix :)  
Definitively gonna check out the RVIZ approach now, had given it some thought already.  

I saw you had some videos out on both the D435 and T265 about performance. You seem to have a very good grasp on them.

Do you have any tips or recommended resources for getting the two of them working together? ODOM from T265 and depth data from D435.

Thank you!
Kristoffer

[matlabbe]

Hi,

On ROS, it can be quite straightforward.
1) Use a static_transform_publisher to publish a TF between the cameras
2) Feed T265 odometry output as input to rtabmap node.
3) Feed color image, depth image and color camera_info from D435 to rtabmap node.

If the frames n the map don't seem correctly aligned together, verify again the TF between the cameras.

cheers,
Mathieu