Simulation

Gazebo

Note

We started once with some initial Panda simulations in Gazebo. However, these simulations are not up-to-date at the moment. So the following explanation should be updated in the future.

Create a catkin workspace

If you want to work with Git version control, then you first need to set up a repository.

Make a catkin workspace in your repository. Some people call their catkin workspace catkin_ws others call it ros_ws. Below you see the example for ros_ws. Go to the terminal and execute the following lines.

cd path/to/your_repository
mkdir -p ros_ws/src
cd ros_ws
source /opt/ros/melodic/setup.sh
catkin_init_workspace src

Visualization

Visualize the Panda robot in RViz

The first thing you want to do with the robot is to visualize the Panda robot in RViz (i.e. Robot Visualization). To do this, we need the kinematics and the meshes of the robot. These two can be found on the Franka Emika's GitHub account.

Create a catkin package called panda_description in the src folder of your ros_ws.

cd path/to/your_repository/ros_ws/src/
catkin_create_pkg panda_description

In your panda_description package:

  • You can compare with my panda_description package I had at the moment I visualized the robot in Rviz. Therefore, go to my repository (constrained_control_robotarm), click on “98 commits”, go to commits on 16 July 2019, click on “view Panda in Rviz”, click on “Browse files”. Now you can see the folders, packages, files I had at that moment.

  • Make a folder called urdf. Copy the urdf and xacro files that are inside the robots folder in franka_ros to your urdf folder. I also did this, but there was no dual_panda example at that moment.

  • Make a folder called meshes. Copy the folders collision and visual from my code into your meshes folder.

  • Make a folder called launch. Copy the two files that are in my launch folder into your launch folder.

  • Check the package.xml and the CMakeLists.txt files. Compare them with the ones I had at that moment and add the lines (e.g. dependencies) you don’t have in your files.

Launch the launch file by entering in the terminal

roslaunch panda_description panda_rviz.launch

Rviz will pop up and will show a Panda robot that is standing still. You can move the joints of this Panda robot by the joint_state_publisher GUI.

todo

Show video.

Visualize the Panda robot in Gazebo

Once you can visualize the Panda robot in RViz, it is time to visualize the robot in Gazebo. Take also a look to this Gazebo tutorial that explains how to use a URDF in Gazebo.

To visualize the Panda robot in Gazebo, we need to include the robot dynamic parameters to the URDF file. Since Franka doesn’t provide the dynamic parameters, I inserted the dynamic parameters from Erdal Pekel.

You can compare with my files I had at the moment I visualized the robot in Gazebo. Therefore, go to my repository (constrained_control_robotarm), click on “98 commits”, go to commits on 17 July 2019, click on “view Panda in Gazebo”, click on “Browse files” . Now you can see the folders, packages, files I had at that moment.

  • Adapt the urdf folder in your panda_description package. (take a look at my files from that moment!)

    • panda_arm_hand.urdf: rigidly fix the base to the Gazebo world

    • hand.xacro: add inertial values

    • panda_arm.xacro: add inertial values + add joint damping

    • panda.gazebo.xacro: new file with gazebo specifications

    • panda_arm_hand.urdf.xacro: include panda.gazebo.xacro

  • Make a catkin package called panda_gazebo in the src folder of your ros_ws.

  • In your panda_gazebo package:

    • Make a folder called worlds. Create a world file including a ground plane, a light source (sun), and a camera at a certain position and orientation

    • Make a folder called launch. Create a launch file.

    • Check the package.xml and the CMakeLists.txt files. Compare them with the ones I had at that moment and add the lines you don’t have.

  • Launch the launch file by entering the terminal

    roslaunch panda_gazebo panda_world.launch

    Gazebo will pop up showing a Panda robot. Despite there being no intentional disturbances in the physics simulator by default, numerical errors should start to build up and cause the Panda robot to move a bit in an uncontrolled way. (In this stage there is no control added to the simulation yet.)

todo

Show video.

Note

In Dynamic Identification of the Franka Emika Panda Robot with Retrieval of Feasible Parameters Using Penalty-Based Optimization and its supplementary material, the authors identified the dynamic parameters of the Panda robot. I implemented them in my code some months ago, check the last version of panda_arm.xacro in panda_description.

ROS Control

To start with the control of the Panda robot in Gazebo, I would also recommend to take a look at the Gazebo ROS Control tutorial.

Position control

  • Adapt the urdf folder in your panda_description package:

    • panda_arm.xacro: add the transmission elements

    • hand.xacro: add the transmission elements

    • panda.gazebo.xacro: add the gazebo_ros_control plugin

  • Adapt two run_depend in the package.xml file of your panda_gazebo package:

    • gazebo_plugins

    • gazebo_ros_control

  • Make a catkin package called panda_control in the src folder of you ros_ws. Once in your panda_control package,

    • make a config folder:

      • add a yaml file called panda_positioncontrol.yaml in which you add a joint_state_controller and joint_position_controllers -> check this file in the current version of my repository: constrained_control_robotarm/ros_ws/src/panda_control/config

    • make a src folder:

      • add a cpp file called panda_positioncontrol_jointspace.cpp in which you make a subscriber (for the current joint states) and a publisher (to publish the reference joint angles) -> check this file in the current version of my repository: constrained_control_robotarm/ros_ws/src/panda_control/src

    • make a launch folder:

      • add a launch file called panda_positioncontrol.launch in which you load the yaml and launch the controller_spawner and the robot_state_publisher -> check this file in the current version of my repository: constrained_control_robotarm/ros_ws/src/panda_control/launch

    • your package.xml file should contain

      • <buildtool_depend>catkin</buildtool_depend>

      • <run_depend>controller_manager</run_depend>

      • <run_depend>joint_state_controller</run_depend>

      • <run_depend>robot_state_publisher</run_depend>

      • <run_depend>effort_controllers</run_depend>

      • <run_depend>roscpp</run_depend>

    • your CMakeLists.txt should contain

      • add_executable(panda_positioncontrol_jointspace src/panda_positioncontrol_jointspace.cpp)

      • target_link_libraries(panda_positioncontrol_jointspace ${catkin_LIBRARIES} )

      • check this file in the current version of my repository: constrained_control_robotarm/ros_ws/src/panda_control, in this stage you will not need: find_package(…), include in include_directories(…), add_library(…), add_dependencies(…), target_link_libraries(…), the other add_executable(…) and target_link_libraries(…)

  • In your panda_gazebo package

    • Add another run_depend in the package.xml file

      • panda_control (the catkin package you just made)

  • in the terminal:

    roslaunch panda_gazebo panda_positioncontrol_jointspace.launch

    • you will be asked to enter a reference in joint space for the Panda arm and a finger displacement for the Panda hand

    • enter these 7 joint angles and the two values for the finger displacement

    • the robot will go to this reference configuration (maybe with some oscillations)

If this works, you can control the Panda robot by publishing joint angles (see your cpp file).

todo

Show video.

Effort control

Control the robot by publishing torques and give task space reference (add inverse kinematics function) instead of joint space reference.

Joint space reference

In the previous step you based yourself on panda_positioncontrol in joint space:

  • panda_positioncontrol.yaml where the effort_controllers/JointPositionController is used (file in panda_control package)

  • panda_positioncontrol_jointspace.cpp (file in panda_control package)

  • panda_positioncontrol.launch (file in panda_control package)

  • you launched it with:

    roslaunch panda_gazebo panda_positioncontrol_jointspace.launch (launch file in panda_gazebo package)

You can also try panda_effortcontrol in joint space, check the following files:

  • panda_effortcontrol.yaml where the effort_controllers/JointEffortController is used (file in panda_control package)

  • panda_effortcontrol_jointspace.cpp (file in panda_control package)

  • panda_effortcontrol.launch (file in panda_control package)

  • you launch it with:

    roslaunch panda_gazebo panda_effortcontrol_jointspace (launch file in panda_gazebo package)

todo

Show video.

Task space reference

You can also try panda_positioncontrol in task space (therefore you need inverse kinematics function, I used the KDL library to do this)

  • panda_positioncontrol.yaml where the effort_controllers/JointPositionController is used (file in panda_control package)

  • panda_positioncontrol_taskspace.cpp (file in panda_control package)

  • panda_positioncontrol.launch (file in panda_control package)

  • you launch it with:

    roslaunch panda_gazebo panda_positioncontrol_taskspace (launch file in panda_gazebo package)

You can also try panda_effortcontrol in task space, check the following files:

  • panda_effortcontrol.yaml where the effort_controllers/JointEffortController is used (file in panda_control package)

  • panda_effortcontrol_taskspace.cpp (file in panda_control package)

  • panda_effortcontrol.launch (file in panda_control package)

  • you launch it with:

    roslaunch panda_gazebo panda_effortcontrol_taskspace (launch file in panda_gazebo package)

todo

Show video.

For the programs in task space, I used the KDL library (since they have an inverse kinematics function). Check my stabilizing_control library for it (in panda_control package in the include folder).

Planning with MoveIt

todo

To write with updated GitHub repo.

ERG

todo

To write with updated GitHub repo.

Planning + ERG

todo

To write with updated GitHub repo.

AGX dynamics

Note

MA2 students have simulated the Panda robot in AGX Dynamics from Algoryx. We will add their tutorial-like explanation in here.

Isaac Sim

Note

We are thinking to use Isaac Sim developed by Nvidia for learning-based control simulations.