Coordinate frame transforms: /odom to /map?

Question

I think I might just be having a difficult time understanding how the coordinate frames relate to each other. I'm working on a visualization (very small subset of rviz) that is shown in a browser.

Basically I'm running SLAM and building a map. Using rosbridge I'm able to get the map and draw it. The base frame is /map and the associated OccupancyGrid has an origin which I translate and rotate by. This seems to work just fine.

Now I want to put an overlay of a robot footprint so you can see where it is, or it thinks it is in the map. Ideally I'd like to display a few topics under /move_base_node which have a base frame of /odom:

• /move_base_node/local_costmap/robot_footprint
• /move_base_node/local_costmap/obstacles
• /move_base_node/local_costmap/inflated_obstacles

** In the end this is what worked for me **

As suggested I look up the transform between map and odom and then publish that on a new topic that I can subscribe to on the client end of rosbridge. Whenever I receive an update I use the transformation between map and odom on any topics with a base frame of odom to bring them into the map frame.

You can check the frames easily with: rostopic echo 'your_topic_name' | grep frame

int main(int argc, char** argv){
  ros::init(argc, argv, "my_tf_listener");
  ros::NodeHandle node;
  ros::Rate rate(1.0);

  tf::TransformListener listener;
  tf::StampedTransform tfTransform;
  tf::Vector3 origin;
  tf::Quaternion rotation;
  tf::Vector3 axis;
  
  static ros::Publisher publisher = 
    node.advertise<geometry_msgs::TransformStamped>("/odom2map_transform",
                            10);

  geometry_msgs::TransformStamped geoTransform;

  int seq = 0;
  geoTransform.header.frame_id = "map";
  geoTransform.child_frame_id = "odom";

  while(node.ok()) {
    try { 
      listener.lookupTransform(geoTransform.header.frame_id,
                   geoTransform.child_frame_id,       
                   ros::Time(0), 
                   tfTransform);
    }
    catch(tf::TransformException &exception) { 
      ROS_ERROR("%s", exception.what());
    }

    origin = tfTransform.getOrigin();
    rotation = tfTransform.getRotation();
    axis = rotation.getAxis();

    geoTransform.header.seq = seq;
    geoTransform.header.stamp = tfTransform.stamp_;

    geoTransform.transform.translation.x = origin.x();
    geoTransform.transform.translation.y = origin.y();
    geoTransform.transform.translation.z = origin.z();

    geoTransform.transform.rotation.x = axis.x();
    geoTransform.transform.rotation.y = axis.y();
    geoTransform.transform.rotation.z = axis.z();
    geoTransform.transform.rotation.w = rotation.getW();

    seq++;
    publisher.publish(geoTransform);
    rate.sleep();
  }
  
  return 0;
}

---

Originally posted by skiesel on ROS Answers with karma: 549 on 2012-08-17

Post score: 0

---

Original comments

Comment by skiesel on 2012-08-18:
I can't figure out where to respond to your comment? robot_footprint for example is in the /odom base frame. I want to figure out how to line up that coordinate frame with the /map frame. Basically I thought I was doing that above? The variables you reference are static, so they're only created once

Comment by Lorenz on 2012-08-19:
Right. Sorry. I missed that... Still I don't see what the sleep is for. I'll update my answer to be more specific.

Answer 1

If you just want to have the pose of the robot in map, all you need to do is look up the transform between /map and your base frame. There is no need to subscribe to the odom topic for that. The code for getting the pose of base_footprint in map would be:

tf::StampedTransform transform_in_map;
try {
  listener.lookupTransform("map", "base_footprint", ros::Time(0), transform_in_map);
} catch(tf::TransformException &exception) {
  ROS_ERROR("%s", exception.what());
}

Most tools, e.g. rviz, are using this mechanism to calculate the robot pose with respect to map instead of directly subscribing to odom. I think the only node that really uses the odom topic on the pr2 is the move_base node because it needs the current velocity. Everything else uses tf.

---

Originally posted by Lorenz with karma: 22731 on 2012-08-18

This answer was ACCEPTED on the original site

Post score: 5

---

Original comments

Comment by pgorczak on 2012-08-23:
The odom topic is also used by mapping and localization algorithms. What can be confusing is that there can be an odometry topic (nav_msgs/Odometry) without the transform being published to tf (for instance if you use robot_pose_ekf that provides the link between /odom and /base)

Comment by Lorenz on 2012-08-23:
According to the ros wiki, neither amcl nor gmapping or hector_mapping subscribe to the odom topic. I also checked the amcl source code, just in case the wiki is wrong. They all use tf, not the odom topic and that's why they work with robot_pose_ekf which publishes the odom->base_link transform.

Comment by pgorczak on 2012-08-23:
Of course - my first sentence didn't make sense...