Differences

This shows you the differences between two versions of the page.

--- teaching:se-kiba:manipulation-assignment:joint-space-controller [2013/05/13 12:56] – [Getting a command inside] bartelsg
+++ teaching:se-kiba:manipulation-assignment:joint-space-controller [2016/05/19 09:19] (current) – external edit 127.0.0.1
@@ Line 8: / Line 8: @@
   * Create a new package for our controllers called iai_seminar_manipulation_controllers. Make sure it depends on the necessary pr2-interface packages.
-  * Create a new h-file for the controller in a suitable include-sub-directory, and add the minimally necessary function headers to comply with the controller interface. Choose a fitting name for the controller. ;)
+  * Copy IAISeminarMultiJointPositionController.h for the controller in a suitable include-sub-directory.
   * Create the corresponding cpp-file in src, and add empty implementations for the functions declared in the h-file.
   * Add our still empty controller as a __regular__ library (NOT a controller plugin library) to the CMakeLists, and try compiling.
@@ Line 72: / Line 72: @@
 ==== Adding a launch-file ====
 We will now add a launch-file to start our controllers automatically:
-  * From the materials-directory, copy the file pr2_multi_joint_position_controllers.yaml into your launch directory.
+  * From the materials-directory, copy the file pr2_multi_joint_position_controllers.yaml into the config directory of our controllers package.
   * Create a new launch-file with a name sounding like it starts the joint position controllers on the pr2.
   * In this launch-file, Load the parameters given in the yaml-file.
@@ Line 106: / Line 106: @@
   * implement the command_callback to copy the new command from the msg inside our controller (using a double buffer and the mutex defined in the h-file)
-  * in the update, always read the current command from the double buffer (also guarded by the mutex), and copy the buffered command into the internal position_command_
+  * in the update, always read the current command from the double buffer (also guarded by the mutex), and copy the buffered command into the internal position_command_. Hint: Implement and use the method copy_from_command_buffer for this. It is supposed to be atomic...
-  *
+  * You can test your code by again echoing the state-topic of the controller and sending a new command from the console:
 <code>
 $ rostopic pub /l_arm_iai_joint_position_controller/command std_msgs/Float64MultiArray '{data: [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6]}'
@@ Line 113: / Line 114: @@
 ==== Using PIDs ====
+For doing the actual control we will use the PID-controller implementation provided by Willow Garage. The h-file already declares a vector of such controllers.
+  * In the init function: Create and push a new PID for every joint to this internal vector.
+  * Also in the init function: read the PID gains from the namespace of the controller. The yaml-file loaded in our launch-file already loads the gains onto the parameter server. Hint: The [[http://ros.org/doc/groovy/api/control_toolbox/html/classcontrol__toolbox_1_1Pid.html|PID class]] has a very useful init-function that reads the gain parameters from a given namespace. Try using it.
+  * Throw a ROS-error if loading of the gains failed. This will warn you of initialization errors.
 ==== Filling in start ====
-==== Doing something meaningful in stop ====
+Now, we will fill the start-hook:
+  * Reset our double buffer for the command to something useful.
+  * Reset the PID (not the gains, though).
+  * Remember the starting time. Hint: You can get it from the [[http://www.ros.org/doc/api/pr2_mechanism_model/html/classpr2__mechanism__model_1_1RobotState.html|robot interface]].
 ==== Closing the loop ====
+Finally, we will calculate the actual feedback control:
+  * get the current time from the robot, calculate some delta-time. and remember the current time for the next control cycle
+  * still: copy in the command from the double buffer
+  * for every joint: calculate the error between actual and commanded position
+  * for every joint: feed the error to the pid-controllers, and command their efforts to the joints of the robot
+  * You can peek again [[http://ros.org/wiki/pr2_mechanism/Tutorials/Writing%20a%20realtime%20joint%20controller|here]].
+  * As a last step: keep on publishing the current state of the controller.
 ==== Using our controller from ====
+TBD.