search

introlab / rtabmap_ros

RTAB-Map's ROS package.
http://wiki.ros.org/rtabmap_ros
BSD 3-Clause "New" or "Revised" License
966 stars 557 forks source link

Interpretation of stats_values #976

Closed sagardhatrak closed 1 year ago

sagardhatrak commented 1 year ago

Hi...

What is interpretation of two stats_values 1. - RtabmapROS/TimeTotal/ms and 2. Timing/Total/ms?? Please explain details about these.

matlabbe commented 1 year ago
sagardhatrak commented 1 year ago

Thank you,

  1. How rtabmap/detection rate is related with Timing/total/ms and Rtabmap/TimeTotal/ms?
  2. For a single value of Timing/total/ms and Rtabmap/TimeTotal/ms, how many frames are assumed to be processed? Is it a single frame? If not, what is the number of frames processed in one bunch of data while calculating both Timings values?
matlabbe commented 1 year ago
  1. If Timing/Total/ms > 1000 and Rtabmap/DetectionRate = 1, the output rate won't be 1 Hz anymore, but less.
  2. Single frame
sagardhatrak commented 1 year ago
  1. Can you please explain detail about first answer? We want to understand, Is there any impact of setting detection rate value on Timing/Total/ms of rtabmap node?
  2. How can output rate be measured in terms of fps?
  3. Can we measure the output rate using rtabmap/MapData? Is it updating and publishing after processing of each frame?
  4. We measured the output rate using RtabmapROS/TimeTotal/ms. we got output rate around 7-8fps with detection rate 30. Why output rate is low? It should be around 30.
  5. RtabmapROS/TimeTotal/ms should be greater than Timing/Total/ms but some time Timing/Total/ms found greater. Why did Timing/Total/ms found to be greater?
  6. Can suggest better way to measure the output rate in terms of fps.
matlabbe commented 1 year ago

  1. Timing/Total/ms will increase accordingly to map size. The impact is that with higher rate, you had more data to the map, thus Timing/Total/ms will increase faster over time.

  2. rostopic hz /rtabmap/info

  3. Yes, you can use rostopic hz /rtabmap/mapData as well.

  4. It means rtabmap cannot run in real-time at this rate. I rarely increase rate over 2 hz.

  5. Are you comparing results in the same message? As shown here: https://github.com/introlab/rtabmap_ros/blob/1a87f1598df2a40d01b230808074b45c9a26039e/rtabmap_slam/src/CoreWrapper.cpp#L2248 The ROS total includes internal total processing time. If it is smaller than Timing/Total/ms, there is something wrong (bug?). Well, we don't use exactly Timing/Total/ms in the cumulative time, but I would assume that the timeRtabmap below estimates the same time than Timing/Total/ms: https://github.com/introlab/rtabmap_ros/blob/1a87f1598df2a40d01b230808074b45c9a26039e/rtabmap_slam/src/CoreWrapper.cpp#L2025-L2028

  6. I am not sure why you want to estimate the FPS of the mapping node. The best way is rostopic hz or you drive it from inverse of RtabmapROS/TimeTotal/ms. For odometry, I would suggest to check the outputs of the odometry node, this one is always trying to process frames as fast as possible.

sagardhatrak commented 1 year ago

Thank you for guidance.

for question 5:

  1. we are comparing the results in the same message.

As per your suggestion, we are going with rostopic hz.