Transformations

In the first part of this day's workshop, we will learn how can we publish, read and visualize transformations.

What is TF?

tf is a package that lets the user keep track of multiple coordinate frames over time.

In this part of the workshop, you will learn how we can use TF in practice and what needs to be taken care of.

Where's the information

The transformation data between the various frames is available across two different topics: /tf_static and /tf.

As the name suggests, the /tf_static topic holds the transformation data between frames that are, well, static! Whereas /tf holds the transformation data between frames which relations change. Let's learn how the two of them differ from one another in practice!

TF static publisher

The static publisher is very useful to publish spatial transformation data that we don't expect to change. In robotics that might be the relation between the robot's base and a fixed point in it's proximity. For example, the transformation between the robot's base and the origin of the table that the robot is mounted on.

Command line demonstration

Open three terminals. In the first terminal, run the following command:

$ rosrun tf2_ros static_transform_publisher 0 0 1 0 0 0 world frame_1

In the second terminal, run the following command:

$ rosrun tf2_ros static_transform_publisher 0 0 2 0 0 0 frame_1 frame_2

Leave these two terminals open and switch to the third.

Let's inspect the content of the /tf_static topic:

$ rostopic echo /tf_static

What do you see? How many messages did you receive? Are they periodic?

Ok, now let's retrieve the transformation between world and frame_2. What do you expect to be the result?

$ rosrun tf tf_echo world frame_2

Note: The tf_echo tool is very useful to evaluate the transformation between two frames on TF.

Python demonstration

Before we proceed we will kill the programs (both static_transform_publishers and tf_echo) that we started before.

Let's get our hands dirty with Python right away. Take the following code snipped and create a script called simple_static_broadcaster.py:

#! /usr/bin/env python

import rospy
import tf2_ros
from geometry_msgs.msg import TransformStamped

if __name__ == '__main__':

    rospy.init_node('tf_static_broadcaster')

    tf_broad = tf2_ros.StaticTransformBroadcaster()

    frame_1 = TransformStamped()
    frame_1.child_frame_id = 'frame_1'
    frame_1.header.frame_id = 'world'

    frame_1.transform.rotation.w = 1.0

    frame_1.transform.translation.z = 1.0
    frame_1.transform.translation.y = 0.0
    frame_1.transform.translation.x = 0.0

    rospy.loginfo("Publishing transform from {0} to {1}".format(frame_1.header.frame_id, frame_1.child_frame_id))
    tf_broad.sendTransform([frame_1])

    rospy.spin()

Run the script. Explore the /tf_static topic again. What do we see?

Note: Do not forget to make it executable with chmod +x simple_static_broadcaster.py.

Intermediate assignment

Write a Python script (name it multiple_frames_static.py) that publishes three frames: world, frame_1 and frame_2. The relations between them should be as follows:

frame_1 is offset by 1m along the Z axis from world
frame_2 is offset by 1m along the X axis from frame_1

Check with tf_echo the relation between world and frame_2. You should see the following output in the command line:

At time 0.000
- Translation: [1.000, 0.000, 1.000]
- Rotation: in Quaternion [0.000, 0.000, 0.000, 1.000]
            in RPY (radian) [0.000, -0.000, 0.000]
            in RPY (degree) [0.000, -0.000, 0.000]

TF publisher

When we want to publish transformation that we know change often, we rather use the "normal" TF publisher. That is, not static.

This type of publisher does not have a command line interface. That's why we will dig into a Python example right away.

Python demonstration

To start with, let's create a copy of the simple_static_broadcaster.py script and call it simple_tf_broadcaster.py. Change the file so that instead of StaticTransformBroadcaster() we use TransformBroadcaster().

After you modified the file run the script. Observe the contents of /tf_static and /tf topic. Seeing any messages?

The reason why you don't see any messages is because TransformBroadcaster() publishes to /tf, which is not a latched topic. Your script published exactly one message to /tf and then got stuck in the rospy.spin(). To confirm that, subscribe to the /tf topic again and start the script, you will see just one message.

Note: You might need to add rospy.sleep(0.1) just before sending the transform. Sometimes, the TransformBroadcaster() needs some time to set everything up.

Let's now write a script, that publishes the frame with a constant frequency:

#! /usr/bin/env python

import rospy
import tf2_ros
from geometry_msgs.msg import TransformStamped

if __name__ == '__main__':

    rospy.init_node('tf_broadcaster')

    tf_broad = tf2_ros.TransformBroadcaster()

    frame_1 = TransformStamped()
    frame_1.child_frame_id = 'frame_1'
    frame_1.header.frame_id = 'world'

    frame_1.transform.rotation.w = 1.0

    frame_1.transform.translation.z = 1.0
    frame_1.transform.translation.y = 0.0
    frame_1.transform.translation.x = 0.0

    while not rospy.is_shutdown():
        tf_broad.sendTransform([frame_1])
        rospy.sleep(0.01)

As you see, we no longer require the rospy.spin(). That is because we use a while loop that does not exit unless we interrupt the execution of the Python script.

Let's inspect the TF topics. As you expect, the /tf_static is not publishing anything. On the other hand, the /tf is publishing a lot of messages:

$ rostopic hz /tf
subscribed to [/tf]
average rate: 95.603
        min: 0.009s max: 0.011s std dev: 0.00031s window: 9

Approximately 100 Hz.

Intermediate assignment

Write a Python script that publishes three frames: world, frame_1 and frame_2. The relations between them should be as follows:

frame_1 is offset by 1m along the Z axis from world
frame_2 is offset by sin(t) along the X and cos(t) along the Z axis from frame_1