ICP Odometry stops output randomly

[masskro0]

ICP Odometry stops output randomly

[matlabbe]

I tried your bag:

ros2 bag play rosbag2_2024_05_28-18_34_14/rosbag2_2024_05_28-18_34_14_0.db3
ros2 run rtabmap_odom icp_odometry --ros-args -r scan:=/sick_tim_left/scan -p frame_id:=sick_tim_left

I don't see odometry getting lost in this one:

One thing I discovered this weekend is that on some machines icp can be super slow. Here the typical performance you should get:

[1717388334.118019026] Odom: ratio=0.680203, std dev=0.008103m|0.002562rad, update time=0.005686s
[1717388334.123897000] Odom: ratio=0.680203, std dev=0.008103m|0.002562rad, update time=0.005381s
[1717388334.185096811] Odom: ratio=0.679901, std dev=0.007922m|0.002505rad, update time=0.005960s
[1717388334.191052915] Odom: ratio=0.679901, std dev=0.007922m|0.002505rad, update time=0.005136s
[1717388334.251740149] Odom: ratio=0.658537, std dev=0.008134m|0.002572rad, update time=0.006070s
[1717388334.258144095] Odom: ratio=0.658537, std dev=0.008134m|0.002572rad, update time=0.005889s
[1717388334.318447767] Odom: ratio=0.681013, std dev=0.007983m|0.002525rad, update time=0.005901s
[1717388334.323697519] Odom: ratio=0.681013, std dev=0.007983m|0.002525rad, update time=0.004749s
[1717388334.385156413] Odom: ratio=0.680798, std dev=0.007840m|0.002479rad, update time=0.005346s

If you have some updates with processing time way higher than other updates, you may try setting export OMP_WAIT_POLICY=passive before launching icp_odometry.

[matlabbe]

Can you share a small rosbag of the scan topic /sick_tim_left/scan till icp_odometry fails? Is the robot navigating in large empty space (out of range of the lidar)?

[masskro0]

rosbag2_2024_05_28-18_34_14.zip
I attached the rosbag. Yes, there are spaces which are out of range. But if that's the problem, then the node should stop immediately after launching, as there are out of range spaces as well.

In case it's relevant, my ros-humble-rtabmap version is 0.21.4-1jammy.20240517.170731 and for ros-humble-rtabmap-odom it's 0.21.4-2jammy.20240522.172901.

[masskro0]

I think it's indeed a hardware performance issue. I couldn't reproduce the issue after setting that flag.

Not related to this issue, but I mentioned that my visual odometry also stops (setting itself to maximum covariance and some random value for yaw velocity). I got that fixed by following this guide for increasing speed/reducing accuracy.

To summarize, it's likely that both odometry nodes stopped due to performance issues.
Thank you for your help!