Robotics Class 2011/Assignment 3

Write a ROS node that subscribes to the "/face_coords" topic that is generated from the previous face detector homework assignment and use the information provided in that topic to move the robot base by publishing messages on the "/cmd_vel" topic. The goal of the assignment is to build a face tracker that attempts to move the robot to keep faces centered in the camera frame. The "/cmd_vel" topic is of type "geometry_msgs/Twist", and you can learn more about it by typing:

rosmsg show geometry_msgs/Twist

For the simulated robot you will use for this assignment, it is important to know its coordinate system. X is positive going forward, Y is positive going to the left, and Z is positive going up. This coordinate system is a right handed coordinate system. We spent a significant amount of time in class going over the importance of keeping track of the coordinate system of the robot, and always making sure you follow the right-hand rule for dealing with the robot coordinate frame.

Because the robot we are using is a differential drive robot, it has two powered wheels and can only move forward and backward, and it can also rotate in place. Thus, for this assignment, when you publish the "/cmd_vel" topic, you need only worry about populating the linear.x (forward and backward motion) and angular.z (turning left and right in place) components of the geometry_msgs/Twist structure. Following the coordinate system above, moving the robot forward would require you to publish a geometry_msgs/Twist message with a positive linear.x component. Rotating the robot in place to the left would require you to publish a geometry_msgs/Twist message with a positive value in the angular.z component. No other fields need be populated in the geometry_msgs/Twist message other than linear.x and angular.z. Remember that you must publish the geometry_msgs/Twist message on the "/cmd_vel" topic. The robot is listening for messages on this topic and will move accordingly.

This assignment is more complicated than the first two, and requires several nodes to be running. First, you must install gazebo and be able to run the simple gazebo empty world launch script. The simple command to start gazebo is:

roslaunch gazebo_worlds empty_world.launch

It may help to follow the ROS instructions here in order to install and get gazebo running.

The next step will be to make sure you have the gazebo_erratic_plugins. These are extensions to gazebo to support differential drive robots (like the iRobot Create) that have two driven wheels. If you installed the ROS installation using Synaptic, you can search Synaptic for "erratic" and you should see a package named ros-diamondback-erratic-robot. You will want to install this package through Synaptic. If you compiled from source, you will want to check out, rosdep install, and rosmake the erratic_robot package.

Once that is complete, you can proceed to checking out the HacDC robot simulation:

cd to where you store your downloaded ROS packages svn co http://hacdc-ros-pkg.googlecode.com/svn/trunk/irobotron_description rosmake irobotron_description

Once that completes, you can start the robot in the simulation (make sure you have done the roslaunch gazebo_worlds... step above before doing this):

roslaunch irobotron_description create_mobile_base.launch

You should then see the robot get inserted into the world. At this point, the robot is up and running in the simulation and you can do a rostopic list to see a variety of message topics. The robot simulator has a camera being simulated that has characteristics similar to the camera on the actual robot. You can subscribe to the simulated robot's camera stream the same way you have done in the past:

rosrun image_view image_view image:=/stereo/left/image_rect

Of course this world is not terribly interesting as it is completely empty. You can add some excitement by using the floating_faces package available in the HacDC ROS repository:

cd to where you store your downloaded ROS packages svn co http://hacdc-ros-pkg.googlecode.com/svn/trunk/floating_faces rosmake floating_faces

Once that is built, you can then launch the floating faces into the world: roslaunch floating_faces faces.launch

Once the faces are in the simulation, it would be useful to be able to manually drive the robot around before trying to control it with a controller. You can do that by checking out the teleop_twist_keyboard ROS package and building it. Once it is built, you can start it by typing:

rosrun teleop_twist_keyboard teleop_twist_keyboard.py

You can following the onscreen instructions on how to use it, but it is advisable to slow the robot commands down by pushing the "z" key a few times, so that the "speed" is around 0.2. This has been found to be a reasonable maximum speed for linear motion with this particular robot. You can move forward by pressing "i", and turn left by pressing "j", but these are all on the on-screen instructions when you run the teleop_twist_keyboard node. At this point, if you are still subscribed to the /stereo/left/image_rect image stream, you should be able to drive around and see what the robot sees, and you should be able to see the faces on the cubes.

Now, you could start up your face detection node from the previous example. Since the simulated robot is publishing the image topic named "/stereo/left/image_rect", your face detection system should work on the faces in the simulated world.

In class we also went over PID control. There is an excellent article on PID here that is recommended reading. Note that for this particular problem, the error term should be thought of as the number of pixels between the center of the detected face and a vertical line running down the center of the image frame. Since the images from the camera are 352 pixels wide, the center of the image is at 176. Then the x coordinate of the "/face_coords" messages can be subtracted from 176 to find the error term that is fed into the controller of your choice to output a commanded velocity that should be fed into the angular.z field of the geometry_msgs/Twist message. Note for this homework, it is recommended to start by only rotating the robot in place to keep an image in the center of the frame (i.e. only populate the angular.z of the geometry_msgs/Twist message). Once you get that working you can think about moving the robot forward and backwards to keep the face at a constant scale, but this will also require modifications to the "/face_coords" message since "/face_coords" currently only provides the center of the located face, not the size of the bounding rectangle. Extra credit is assigned for getting these modifications into your solution.

There is a complete solution to the homework assignment if you are interested in studying it. It can be checked out from the ROS HacDC repository. There are two packages. The first is a generic PID control library and the second is the controller itself.

cd to where you store your downloaded ROS packages svn co http://hacdc-ros-pkg.googlecode.com/svn/trunk/pid_control rosmake pid_control

cd to where you store your downloaded ROS packages svn co http://hacdc-ros-pkg.googlecode.com/svn/trunk/face_follow rosmake face_follow

To run the face tracker, you can launch it as follows: roslaunch face_follow follow.launch

Don't forget for it to work you need to start a face detector first, either your own from the last homework, or the example solution from the last assignment.

Back to Robotics Class 2011