Large point cloud messages

[Rodolfo]

I'm having trouble with the network handling full Velodyne 16 pointcloud messages. I can limit the scan to just the forward 180 degrees of the circle. How do I change parameters in Rtabmap to properly work this change? (Or is there a better solution?)

[matlabbe]

There are no particular parameters about the shape of the point cloud, either it is 180 deg or 360 deg. As long as the timestamps of each point of the scan are provided, deskewing should still work.

What is the maximum bandwidth that you can use? Why do you need to stream over network? Other solutions would be to process the scans onboard.

[Rodolfo]

Thanks for this. I don't know that the Raspberry Pi 5 (8G mem) would be powerful enough to run Rtabmap. Maybe it would? That's why I was sending the point clouds to the laptop.

[matlabbe]

Are you using icp_odometry? If so, you could probably use only that node onboard, at worst just increase Icp/VoxelSize, OdomF2M/ScanSubtractRadius and Icp/MaxCorrespondenceDistance (roughly 10x the voxel size) parameters.

You could still run rtabmap node remotely, but would only need 1 Hz scans instead of the frame rate your are using.

[Rodolfo]

Good! Is it better to use ICP odometry than the odometry from the Oak camera images? I see that the IMU data comes at a high rate from the Oak camera, also. Last night I found that the Raspi 5 could process the slam from the camera images successfully. I also ran the Velodyne but didn't work with its output yet on the robot. On the laptop I only ran the joystick and Rtabmap-viz.

[matlabbe]

You may compare each other and use the one drifting less. If you are using the full 360 FOV of the lidar, it is generally more robust than single-stereo camera indoor.