Ask Your Question

TF updating very slowly

asked 2015-02-16 01:28:18 -0500

Robocop87 gravatar image

updated 2015-02-16 02:58:09 -0500

Hey everyone,

I am trying to publish Pose messages based on distances I get from ar_track_alvar. I have static transforms which relate the markers to the map. These static transforms broadcast at 100Hz. I take the position and orientation data from the marker message and create a transform from it. I then broadcast that transform (I can't use the one supplied by ar_track_alvar so I create my own). I then listen for a transform from the map to my robot in order to get my position and orientation with respect to the map, and then publish that as a pose message. This works fine, but the problem is that the poses are published with an intolerable delay. After moving my robot, it can take up to a minute for the pose to update. To be clear, poses continue to be published, but they reflect the old position of the data. I used rosrun tf tf_echoto track the tfs, and I found that the tf is reflecting old data also.

Additionally, I see a lot of errors saying:

Lookup would require extrapolation into the past. Requested time XXXXXX but the earliest data is at time YYYYYYY, when looking up transform from frame [thor] to frame [map].

Despite my constant broadcasting and the static transforms.

Things I have tried:

I have gotten rid of the waitForTransform call in case that was adding some latency. I have reduced the tag subscribers buffer to 1 to ensure processing of only the most recent message. I have added extra AsyncSpinners.

These had no effects. Does anyone have any idea why it is taking the tfs so long to broadcast?

Relevant code:

//Figure out which tag we are looking at.
    int tagID = msg->markers[0].id;
    std::stringstream ss;
    ss << "hsmrs/marker_" << tagID;
    std::string markerFrameID = ss.str();

    //Manually create transform from robot to tag
    tf::Transform transform;

    double x, y;
    x = msg->markers[0].pose.pose.position.x;
    y = msg->markers[0].pose.pose.position.y;

    tf::Quaternion q;
    tf::quaternionMsgToTF(msg->markers[0].pose.pose.orientation, q);

    transform.setOrigin( tf::Vector3(x, y, 0.0) );

    //Broadcast transform
    ROS_INFO("Broadcasting transform!");
    br.sendTransform(tf::StampedTransform(transform, ros::Time::now(), markerFrameID, "thor"));
    ROS_INFO("Transform broadcasted!");

    //Listen for transform from map to robot
    ROS_INFO("Listening for transform!");
    tf::StampedTransform mapTransform;
                listener.waitForTransform("map", "thor", ros::Time(0) ros::Duration(0.1));
        listener.lookupTransform("map", "thor",
            ros::Time(0), mapTransform);
        ROS_INFO("Transform heard");    
    catch (tf::TransformException &ex) {

    //Unpack position and orientation
    double map_x, map_y;
    map_x = mapTransform.getOrigin().x();
    map_y = mapTransform.getOrigin().y();

    geometry_msgs::Quaternion quat;
    tf::quaternionTFToMsg(mapTransform.getRotation(), quat);

    //Publish pose as geometry message
    geometry_msgs::Pose pose;
    pose.position.x = map_x;
    pose.position.y = map_y;
    pose.orientation = quat;



The static publisher was running on a separate computer. By running it on the robot this problem disappeared. But I need it to work on the workstation computer. How can I make it work?

edit retag flag offensive close merge delete

2 Answers

Sort by » oldest newest most voted

answered 2015-02-16 03:10:59 -0500

gvdhoorn gravatar image

updated 2015-02-16 03:14:26 -0500


The static publisher was running on a separate computer. By running it on the robot this problem disappeared. But I need it to work on the workstation computer. How can I make it work?

Have you checked the clocks on both computers? TF has a temporal aspect to it, and if the clocks are not in sync, things start to go wrong.

From - Introduction:

When using wall-clock time on multiple machines, it is important that you synchronize time between them. ROS does not provide this functionality as there are already well-established methods (e.g. ntp) for doing this. If you do not synchronize the wall-clocks of multiple machines, they will not agree on temporal calculations like those used in tf.

See also - Timing issues, TF complaining about extrapolation into the future?, and Why robot needs to set NTP? fi.

edit flag offensive delete link more


I'll look into this, that may very well be the problem!

Robocop87 gravatar image Robocop87  ( 2015-02-16 10:25:42 -0500 )edit

answered 2015-02-16 04:00:03 -0500

Procópio gravatar image

updated 2015-02-16 04:02:40 -0500

how are you setting up your ROS_IP and ROS_HOSTNAME? Are you using an IP or a name to be resolved? In my experience, I had better results using directly the IPs numbers, you can give it a try. you can always review your network setup

edit flag offensive delete link more


I am using direct IP addresses, no names.

Robocop87 gravatar image Robocop87  ( 2015-02-16 10:26:04 -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: 2015-02-16 01:28:18 -0500

Seen: 1,681 times

Last updated: Feb 16 '15