Warning: TF_REPEATED_DATA ignoring data with redundant timestamp for frame odom_comb

[fjp]

Hi, when working with amcl for example the mir-robot Gazebo demo (existing map) a lot of the following warnings are logged:

Warning: TF_REPEATED_DATA ignoring data with redundant timestamp for frame odom_comb at time 23.567000 according to authority unknown_publisher
         at line 278 in /tmp/binarydeb/ros-noetic-tf2-0.7.5/src/buffer_core.cpp

I guess this is related to ros/geometry2#467. To reduce the number of TF_REPEATED_DATA warnings the laser scan publish rate needs to be reduced, because amcl depends on rate of incoming laser messages.

According to the discussion here this seems to be a bug in amcl:

Basically, once amcl has published a TF for a given time stamp t, it must never publish it again. So there are two options: Either publish the first message and then drop all subsequent ones, or delay publishing the message until amcl is certain that this is the definitive one. Both have their advantages and disadvantages.

Can you provide some more hints on how to fix this warning? I assume the problem is in laserReceived()?

[mikeferguson]

I don't actually see how it would be possible for AMCL to publish two transforms with the same timestamp unless it receives two laser scans with the same timestamp - as far as I can see, the only place we publish transforms is inside laserReceived() - and we use the timestamp of the laser message received. To me that indicates either an issue with the Gazebo laser plugin, or some issue in ROS that is causing the same message to be delivered to AMCL twice.

[fjp]

I can confirm that the warning log messages can be reduced when the update rate of the laser is reduced. Probably there is an issue in the gazebo_ros_gpu_laser plugin that I am using or maybe there is an issue in gazebo and its sensor update behaviour.

[reinzor]

When using gazebo without the lockstep parameter, the lidar render thread and the physics thread are not synchronized. When your graphics card can easily keep up and you set your physics update rate low, you are rendering multiple laser scan messages per physics tick which could result in rendering and publishing messages with the same timestamps in the gazebo_ros_gpu_laser plugin. If you start gazebo with --lockstep you tell gazebo that the sensors should respect the update rates that are specified:

  --lockstep                    Lockstep simulation so sensor update rates are respected.

[fjp]

Thanks @reinzor this looks promising. However, I still see the TF warnings being published afer I have added the lockstep in the mir_empty_world.launch:

      <include file="$(find gazebo_ros)/launch/empty_world.launch">
        <arg name="world_name" value="$(arg world_name)"/>
        <arg name="paused" value="true" />
        <arg name="gui" value="$(arg gui)" />
        <arg name="lockstep" default="true" />
      </include>

I am not sure if I use the lockstep parameter correctly in the launch file.

[reinzor]

The empty_world.launch launch file in the gazebo_ros package does not consume this parameter:

$ roslaunch gazebo_ros empty_world.launch --ros-args
Required Arguments:
  command_arg1: undocumented
  command_arg2: undocumented
  command_arg3: undocumented
  extra_gazebo_args: undocumented
Optional Arguments:
  debug (default "false"): undocumented
  enable_ros_network (default "true"): undocumented
  gui (default "true"): undocumented
  gui_required (default "false"): undocumented
  headless (default "false"): undocumented
  output (default "screen"): undocumented
  paused (default "false"): undocumented
  physics (default "ode"): undocumented
  pub_clock_frequency (default "100"): undocumented
  recording (default "false"): undocumented
  respawn_gazebo (default "false"): undocumented
  server_required (default "false"): undocumented
  use_clock_frequency (default "false"): undocumented
  use_sim_time (default "true"): undocumented
  verbose (default "false"): undocumented
  world_name (default "worlds/empty.world"): undocumented

The gzserver executable in the gazebo_ros package does:

$ rosrun gazebo_ros gzserver --help
 --help -s /opt/ros/noetic/lib/libgazebo_ros_paths_plugin.so -s /opt/ros/noetic/lib/libgazebo_ros_api_plugin.so
gzserver -- Run the Gazebo server.

`gzserver` [options] <world_file>

Gazebo server runs simulation and handles commandline options, starts a Master, runs World update and sensor generation loops.

Options:
  -v [ --version ]              Output version information.
  --verbose                     Increase the messages written to the terminal.
  -h [ --help ]                 Produce this help message.
  -u [ --pause ]                Start the server in a paused state.
  --lockstep                    Lockstep simulation so sensor update rates are 
                                respected.
  -e [ --physics ] arg          Specify a physics engine 
                                (ode|bullet|dart|simbody).
  -p [ --play ] arg             Play a log file.
  -r [ --record ]               Record state data.
  --record_encoding arg (=zlib) Compression encoding format for log data 
                                (zlib|bz2|txt).
  --record_path arg             Absolute path in which to store state data
  --record_period arg (=-1)     Recording period (seconds).
  --record_filter arg           Recording filter (supports wildcard and regular
                                expression).
  --record_resources            Recording with model meshes and materials.
  --seed arg                    Start with a given random number seed.
  --iters arg                   Number of iterations to simulate.
  --minimal_comms               Reduce the TCP/IP traffic output by gzserver
  -s [ --server-plugin ] arg    Load a plugin.
  -o [ --profile ] arg          Physics preset profile name from the options in
                                the world file.

You could use the extra_gazebo_args parameter in the launch file here.

[brean]

I am having the same problem. I tried setting --lockstep for empty_world but I am still getting the log filled up with the warning.

<include file="$(find gazebo_ros)/launch/empty_world.launch">
[...]
    <arg name="extra_gazebo_args" value="--lockstep"/>
</include>

Any other ideas?

[kvnptl]

This is not an answer to the question, but if you just want to avoid/hide the warnings and analyze the code output on terminal, then you can use the below command at the end of the original command:

2> >(grep -v TF_REPEATED_DATA buffer_core)

Reference: #467

[Daviesss]

I solved this problem from source . The source code of amcl . To reduce the number of TF_REPEATED_DATA warnings the laser scan publish rate needs to be reduced, because amcl depends on rate of incoming laser messages. I did this in the .cpp node and it totally worked fine for me . If anyone has this issue , you can follow this procedure.

[brean]

@Daviesss sounds like we want to configure this as parameter instead of setting it hardcoded in the c++-file. Maybe update your code a bit and create a pull-request for AMCL?

[Daviesss]

@Daviesss sounds like we want to configure this as parameter instead of setting it hardcoded in the c++-file. Maybe update your code a bit and create a pull-request for AMCL?

I don't mind sending a pull request , will do that before the week ends.

Warm regards,

Davies.

[Daviesss]

@Daviesss sounds like we want to configure this as parameter instead of setting it hardcoded in the c++-file. Maybe update your code a bit and create a pull-request for AMCL?

Sent in a pull request #1260

Warm regards.