Ask Your Question

TF behaviour showing delay when using ros::Time::now() vs. ros::Time(0)

asked 2017-05-23 02:26:36 -0500

danyloM gravatar image

updated 2017-05-24 04:14:43 -0500

My situation is that I have /tf topic publishing the following frame transforms:

  • A to B
  • B to C
  • C to D

My goal is to create a simple C++ program which extracts the transform A to D at the current time (i.e. ros::Time::now()). I call E the extracted target frame, and ideally E should coincide with D at each 10 Hz update of the while loop that I am using to listen in on the /tf topic. Here is the C++ code:

#include <ros/ros.h>
#include <tf/transform_listener.h>
#include <tf/transform_broadcaster.h>

int main(int argc, char** argv)
  ros::init(argc, argv, "tf_measurement_jumping_debug");
  ros::NodeHandle nh;

  tf::TransformListener tf_listener;
  tf::TransformBroadcaster tf_broadcaster;
  tf::StampedTransform transform;

  ros::Rate loop_rate(10);
  while (ros::ok())
      // Get the current transform from A to D
      ros::Time tf_time = ros::Time::now();
      tf_listener.waitForTransform("A", "D", tf_time, ros::Duration(3.0));
      tf_listener.lookupTransform("A", "D", tf_time, transform);
      // Publish back what I got
      transform.child_frame_id_ = "E"; // Rename to not overwrite D
    catch (tf::TransformException ex)

The result is not satisfactory as shown in the following screen recording:

image description

There is some kind fo delay in the transform from frame A to E. The reason I think that this is a delay issue is because if I move frame B slowly enough, then E more or less coincides with D (note that the transform B to C is rigid).

In the above code, if instead of ros::Time tf_time = ros::Time::now(); I use ros::Time tf_time = ros::Time(0); then the result is what I want:

image description

My question is what is the cause of this? Ideally, I want a solution using ros::Time::now() because

  1. What I am looking for is the current transform, not the latest transform;
  2. When transforms stop arriving (e.g. because transform C to D is no longer being published) then the ros::Time(0) "hack" will keep sending the last transform prior to this event and I do not want that.

Thank you very much for your help. I have looked at the /tf tutorials and saw the timestamp debugging section - however unfortunately none of it has so far helped me to resolve this problem.

Update (tf_monitor output)

Here's the output of rosrun tf tf_monitor:

RESULTS: for all Frames

Frame: B published by unknown_publisher Average Delay: 0.000125737 Max Delay: 0.0130457
Frame: C published by unknown_publisher Average Delay: 0.00024707 Max Delay: 0.0130457
Frame: D published by unknown_publisher Average Delay: 0.24257 Max Delay: 0.282611

All Broadcasters:
Node: unknown_publisher 196.535 Hz, Average Delay: 0.00547737 Max Delay: 0.276091

And here is the output of rosrun tf tf_monitor A D:

RESULTS: for A to D
Chain is: A -> B -> C -> D
Net delay     avg = 0.864727: max = 3.39458

Frame: B published by unknown_publisher Average Delay: -2.90574e-05 Max Delay: 0.0139616
Frame: C published by unknown_publisher Average Delay: 0.000132843 ...
edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted

answered 2017-05-23 03:19:58 -0500

fvd gravatar image

Have you checked which transforms are published at which time with rosrun tf tf_monitor? When you call waitForTransform, I believe it waits for all of the transforms to be republished and this introduces some unnecessary delay.

You probably want to update E when even one of the transforms is new, so using ros::Time(0) is preferable until one of the transforms stops arriving. You should be able to check with waitForTransform("A", "D", ros::Time::now()-0.5, ros::Duration(3.0)), which should return false if the necessary transforms have been missing for more than 0.5 seconds. I'm not sure about the syntax though.

edit flag offensive delete link more


I added an update to my original post with the output of tf_monitor. Indeed, I want to update E when even one of the transforms is new (e.g. the A to B transform, which updates much faster than the C to D transform). Does the tf_monitor output give any more clues?

danyloM gravatar imagedanyloM ( 2017-05-24 04:16:29 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools



Asked: 2017-05-23 02:26:36 -0500

Seen: 998 times

Last updated: May 24 '17