This project is an implementation of Cartesian impedance control for robotic manipulators. It is a type of control strategy that sets a dynamic relationship between contact forces and the position of a robot arm, making it suitable for collaborative robots. It is particularily useful when the interesting dimensions in the workspace are in the Cartesian space.
The controller is developed using the seven degree-of-freedom (DoF) robot arm LBR iiwa
by KUKA AG
and has also been tested with the Franka Emika Robot (Panda)
both in reality and simulation.
This controller is used and tested with ROS 1 melodic
and noetic
.
The implementation consists of a
- base library that has few dependencies and can e.g. be directly integrated into software such as the DART simulator and a
- ROS control integration on top of it.
http://www.youtube.com/watch?v=Q4aPm4O_9fY
- Configurable stiffness values along all Cartesian dimensions at runtime
- Configurable damping factors along all Cartesian dimensions at runtime
- Change reference pose at runtime
- Apply Cartesian forces and torques at runtime
- Optional filtering of stiffnesses, pose and wrenches for smoother operation
- Handling of joint trajectories with nullspace configurations, e.g. from MoveIt
- Jerk limitation
- Separate base library that can be integrated in non-ROS environments
- Interface to ROS messages and dynamic_reconfigure for easy runtime configuration
The torque signal commanded to the joints of the robot is composed by the superposition of three joint-torque signals:
- The torque calculated for Cartesian impedance control with respect to a Cartesian pose reference in the frame of the EE of the robot (
tau_task
). - The torque calculated for joint impedance control with respect to a desired configuration and projected in the nullspace of the robot's Jacobian, so it should not affect the Cartesian motion of the robot's end-effector (
tau_ns
). - The torque necessary to achieve the desired external force command (
cartesian_wrench
), in the frame of the EE of the robot (tau_ext
).
- Joint friction is not accounted for
- Stiffness and damping values along the Cartesian dimensions are uncoupled
- No built-in gravity compensation for tools or workpieces (can be achieved by commanding a wrench)
We use RBDyn
to calculate forward kinematics and the Jacobian.
The installation steps for the installation of the non-ROS dependencies are automated in scripts/install_dependencies.sh
.
Assuming that there is an initialized catkin workspace you can clone this repository, install the dependencies and compile the controller.
Here are the steps:
cd catkin_ws
git clone https://github.com/matthias-mayr/Cartesian-Impedance-Controller src/Cartesian-Impedance-Controller
src/Cartesian-Impedance-Controller/scripts/install_dependencies.sh
rosdep install --from-paths src --ignore-src --rosdistro=${ROS_DISTRO} -y
catkin_make # or 'catkin build'
source devel/setup.bash
This allows you to add a controller configuration for the controller type cartesian_impedance_controller/CartesianImpedanceController
in your ros_control
configuration.
When using the controller it is a good practice to describe the parameters in a YAML
file and load it. Usually this is already done by your robot setup - e.g. for iiwa_ros that is here.
Here is a template of what needs to be in that YAML file that can be adapted:
CartesianImpedance_trajectory_controller:
type: cartesian_impedance_controller/CartesianImpedanceController
joints: # Joints to control
- iiwa_joint_1
- iiwa_joint_2
- iiwa_joint_3
- iiwa_joint_4
- iiwa_joint_5
- iiwa_joint_6
- iiwa_joint_7
end_effector: iiwa_link_ee # Link to control arm in
update_frequency: 500 # Controller update frequency in Hz
# Optional parameters - the mentioned values are the defaults
dynamic_reconfigure: true # Starts dynamic reconfigure server
handle_trajectories: true # Accept traj., e.g. from MoveIt
robot_description: /robot_description # In case of a varying name
wrench_ee_frame: iiwa_link_ee # Default frame for wrench commands
delta_tau_max: 1.0 # Max. commanded torque diff between steps in Nm
filtering: # Update existing values (0.0 1.0] per s
nullspace_config: 0.1 # Nullspace configuration filtering
pose: 0.1 # Reference pose filtering
stiffness: 0.1 # Cartesian and nullspace stiffness
wrench: 0.1 # Commanded torque
verbosity:
verbose_print: false # Enables additional prints
state_msgs: false # Messages of controller state
tf_frames: false # Extra tf frames
To start up with this controller, eventually the controller spawner needs to load the controller. Typically this is baked into the robot driver. For example if using the YAML example above, with iiwa_ros, this can be achieved with command:
roslaunch iiwa_gazebo iiwa_gazebo.launch controller:=CartesianImpedance_trajectory_controller
If it is not deactivated, the controller can be configured with dynamic_reconfigure both with command line tools as well as the graphical user interface rqt_reconfigure. To start the latter you can run:
rosrun rqt_reconfigure rqt_reconfigure
There are several entries:
cartesian_wrench_reconfigure
damping_factors_reconfigure
stiffness_reconfigure
For applying wrench, the apply
checkbox needs to be ticked for the values to be used. Damping and stiffness changes are only updated when the update
checkbox is ticked, allowing to configure changes before applying them. Note that the end-effector reference pose can not be set since it usually should follow a smooth trajectory.
In addition to the configuration with dynamic_reconfigure
, the controller configuration can always be adapted by sending ROS messages. Outside prototyping this is the main way to parameterize it.
The instructions below use ${controller_ns}
for the namespace of your controller. This can for example be /cartesian_impedance_controller
. If you want to copy the example commands 1:1, you can set an environment variable with export controller_ns=/<your_namespace>
.
New reference poses can be sent to the topic ${controller_ns}/reference_pose
. They are expected to be in the root frame of the robot description which is often world
. The root frame can be obtained from the parameter server with rosparam get ${controller_ns}/root_frame
.
To send a new reference pose 0.6m above the root frame pointing into the z-direction of it, execute this:
rostopic pub --once /${controller_ns}/reference_pose geometry_msgs/PoseStamped "header:
seq: 0
stamp:
secs: 0
nsecs: 0
frame_id: ''
pose:
position:
x: 0.0
y: 0.0
z: 0.6
orientation:
x: 0.0
y: 0.0
z: 0.0
w: 1.0"
Most often one wants to have a controlled way to set reference poses. Once can for example use a trajectory generator for Cartesian trajectories.
In order to set only the Cartesian stiffnesses, once can send a geometry_msgs/WrenchStamped
to set_cartesian_stiffness
:
rostopic pub --once /${controller_ns}/set_cartesian_stiffness geometry_msgs/WrenchStamped "header:
seq: 0
stamp:
secs: 0
nsecs: 0
frame_id: ''
wrench:
force:
x: 300.0
y: 300.0
z: 300.0
torque:
x: 30.0
y: 30.0
z: 30.0"
The damping factors can be configured with a geometry_msgs/WrenchStamped
msg similar to the stiffnesses to the topic ${controller_ns}/set_damping_factors
. Damping factors are in the interval [0.01,2]. These damping factors are additionally applied to the damping rule which is 2*sqrt(stiffness)
.
When also setting the nullspace stiffnes, a custom messages of the type cartesian_impedance_controller/ControllerConfig
is to be sent to set_config
:
rostopic pub --once /${controller_ns}/set_config cartesian_impedance_controller/ControllerConfig "cartesian_stiffness:
force: {x: 300.0, y: 300.0, z: 300.0}
torque: {x: 30.0, y: 30.0, z: 30.0}
cartesian_damping_factors:
force: {x: 1.0, y: 1.0, z: 1.0}
torque: {x: 1.0, y: 1.0, z: 1.0}
nullspace_stiffness: 10.0
nullspace_damping_factor: 0.1
q_d_nullspace: [0, 0, 0, 0, 0, 0, 0]"
q_d_nullspace
is the nullspace joint configuration, so the joint configuration that should be achieved if the nullspace allows it. Note that the end-effector pose would deviate if the forward kinematics of this joint configuration do not overlap with it.
A Cartesian wrench can be commanded by sending a geometry_msgs/WrenchStamped
to the topic ${controller_ns}/set_cartesian_wrench
.
Internally the wrenches are applied in the root frame. Therefore wrench messages are transformed into the root frame using tf
.
Note: An empty field frame_id
is interpreted as end-effector frame since this is the most applicable one when working with a manipulator.
Note: The wrenches are transformed into the root frame when they arrive, but not after that. E.g. end-effector wrenches might need to get updated.
rostopic pub --once /${controller_ns}/set_cartesian_wrench geometry_msgs/WrenchStamped "header:
seq: 0
stamp:
secs: 0
nsecs: 0
frame_id: ''
wrench:
force:
x: 0.0
y: 0.0
z: 5.0
torque:
x: 0.0
y: 0.0
z: 0.0"
If handle_trajectories
is not disabled, the controller can also execute trajectories. An action server is spun up at ${controller_ns}/follow_joint_trajectory
and a fire-and-forget topic for the message type trajectory_msgs/JointTrajectory
is at ${controller_ns}/joint_trajectory
.
In order to use it with MoveIt
its controller configuration (example in iiwa_ros) needs to look somewhat like this:
controller_list:
- name: ${controller_ns}
action_ns: follow_joint_trajectory
type: FollowJointTrajectory
default: true
joints:
- iiwa_joint_1
- iiwa_joint_2
- iiwa_joint_3
- iiwa_joint_4
- iiwa_joint_5
- iiwa_joint_6
- iiwa_joint_7
Note: A nullspace stiffness needs to be specified so that the arm also follows the joint configuration and not just the end-effector pose.
We have used the controller with Cartesian translational stiffnesses of up to 1000 N/m and experienced it as very stable. It is also stable in singularities.
One additional measure can be to limit the maximum joint torques that can be applied by the robot arm in the URDF. On our KUKA iiwas we limit the maximum torque of each joint to 20 Nm, which allows a human operator to easily interfere at any time just by grabbing the arm and moving it.
When using iiwa_ros
, these limits can be applied here. For the Panda they are applied here. Both arms automatically apply gravity compensation, the limits are only used for the task-level torques on top of that.
The source code comes with Doxygen documentation. In a catkin
workspace it can be built with:
sudo apt-get install ros-$ROS_DISTRO-rosdoc-lite
roscd cartesian_impedance_controller
rosdoc_lite .
It can then be found in the doc
folder with doc/html/index.html
being the entry point.
The documentation for the public Github repository is automatically built and is deployed at:
https://matthias-mayr.github.io/Cartesian-Impedance-Controller/
The main public code repository is at: https://github.com/matthias-mayr/Cartesian-Impedance-Controller
Issues, questions and pull requests are welcome. Feel free to contact the main author at [email protected]
if you are using this implementation.
A brief paper about the features and the control theory is accepted at JOSS. If you are using it or interacting with it, we would appreciate a citation:
Mayr et al., (2024). A C++ Implementation of a Cartesian Impedance Controller for Robotic Manipulators. Journal of Open Source Software, 9(93), 5194, https://doi.org/10.21105/joss.05194
@article{mayr2024cartesian,
doi = {10.21105/joss.05194},
url = {https://doi.org/10.21105/joss.05194},
year = {2024},
publisher = {The Open Journal},
volume = {9},
number = {93},
pages = {5194},
author = {Matthias Mayr and Julian M. Salt-Ducaju},
title = {A C++ Implementation of a Cartesian Impedance Controller for Robotic Manipulators}, journal = {Journal of Open Source Software}
}
catkin build shows this CMake Error:
CMake Error at /usr/share/cmake-3.16/Modules/FindPkgConfig.cmake:463 (message):
A required package was not found
Call Stack (most recent call first):
/usr/share/cmake-3.16/Modules/FindPkgConfig.cmake:643 (_pkg_check_modules_internal)
CMakeLists.txt:12 (pkg_check_modules)
There are missing dependencies. When replacing catkin_ws
with your workspace, they can be resolved like this :
cd catkin_ws
src/cartesian_impedance_controller/scripts/install_dependencies.sh
rosdep install --from-paths src --ignore-src --rosdistro=${ROS_DISTRO} -y
When starting the controller, this message appears:
[ControllerManager::loadController]: Could not load class 'cartesian_impedance_controller/CartesianImpedanceController': Failed to load library /home/matthias/catkin_ws/devel/lib//libcartesian_impedance_controller_ros.so. Make sure that you are calling the PLUGINLIB_EXPORT_CLASS macro in the library code, and that names are consistent between this macro and your XML. Error string: Could not load library (Pocoexception = libRBDyn.so.1: cannot open shared object file: No such file or directory) /gazebo: [ControllerManager::loadController]: Could not load controller 'CartesianImpedance_trajectory_controller' because controller type 'cartesian_impedance_controller/CartesianImpedanceController' does exist.
This happens when a shared library can not be found. They are installed with scripts/install_dependencies.sh
. The dynamic linker has a cache and we now manually update it by calling ldconfig
after the installation.
Most likely this happens because some parts of the stack like iiwa_ros
or RBDyn
were built with SIMD
/march=native
being turned on and other parts are not. Everything needs to be built with or without this option in order to have working alignment. This package builds without, because it is otherwise cumbersome for people to ensure that this happens across the whole stack.