Documentation Status

Cannot load information on name: ee_cart_imped_control, distro: groovy, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: ee_cart_imped_control, distro: hydro, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: ee_cart_imped_control, distro: indigo, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.
Cannot load information on name: ee_cart_imped_control, distro: jade, which means that it is not yet in our index. Please see this page for information on how to submit your repository to our index.

Overview

The ee_cart_imped_control (for End Effector Cartesian Impedance Control) implements a controller that executes trajectories for force/impedance control of the end effector of a chain of joints. Each point on the trajectory specifies a wrench or a stiffness around a Cartesian degree of freedom as well as a Cartesian position and orientation. For a degree of freedom, if a stiffness is specified, the controller will attempt to reach the position given with that stiffness (mimicking a proportional position controller). If a wrench is specified, the position information is ignored.

The controller is an open-loop Jacobian-transpose force controller. At every point on the trajectory the desired stiffness or desired wrench is converted into a Cartesian force/wrench vector and ultimately into a joint torque vector using the transpose of the arm's instantaneous Jacobian matrix. No guarantee is made to the magnitude of the applied force/wrench/impedance because this is an open-loop controller. There is currently not adequate feedback on the arm to determine the applied force/wrench/impedance. Joint stiction, joint position limits, motor torque limits, and sometimes the arm configuration (at a singularity, for example) will cause the applied force/wrench/impedance to deviate from the desired values. Generally, unless the arm is at a singularity or a joint is at its limit, the direction of the force/wrench/impedance will be correct.

The controller takes trajectory points and linearly interpolates, in Cartesian space, between positions and orientations. However, it does not interpolate between force or stiffness, but simply holds the requested force or stiffness constant for each segment.

This controller should never be used directly; rather, all interaction with it should take place through the ee_cart_imped_action.

Videos

In this video, the PR2 first moves its gripper to the center of its body and then applies a constant force in the x direction (towards/away from the robot). As you can see, the force is light so the human can easily stop the motion. However, with no intervention, the robot hand will continue to move as far in the x direction as it can go. There is no sense of trying to achieve a goal in this direction. This type of control is useful if you wish to exert a constant force against something.

In this video, the PR2 first moves its gripper to the center of its body and then tries to extend the gripper about 25 cm farther away, but with low stiffness in the x direction (towards/away from the robot). Since the stiffness is low, it is easy for a human to move the hand away from the center of the body in the x direction and it returns slowly. In the y and z directions, however, where the stiffness is set to the maximum, it takes a lot of strength to displace the gripper and it returns very quickly to its set point.

This controller is useful for any activity in which it is necessary to use less than the full strength of the arms. This video shows several short clips of the PR2 using the force/impedance controller for various tasks.

For even more videos, see the CSAIL ee_cart_imped videos.

ROS API

Parameters

Parameters

root_name (string)
  • The base frame in which the end effector motion will be specified.
tip_name (string)
  • The frame of the tip

Example configuration (from ee_cart_imped_control/pr2_arms_cart_imped_controller.yaml):

r_arm_cart_imped_controller:
  type: EECartImpedControlPlugin
  root_name: torso_lift_link
  tip_name: r_gripper_tool_frame

Topics

Published Topics

state (ee_cart_imped_msgs/EECartImpedFeedback)
  • Current state of the controller, including the current and desired positions and orientations, the desired force or stiffness, and the force or stiffness passed to the joints (TODO: Make this the actual joint effort in Cartesian space).

Subscribed Topics

command (ee_cart_imped_msgs/EECartImpedGoal)
  • The trajectory to follow. The absolute start time of the trajectory is given in header.stamp. If forces are specified, the controller will try to exert that force in the correct direction throughout the trajectory segment. If stiffness is specified, the controller will try to achieve the desired segment point with that stiffness. At the end of the trajectory, the controller holds the last point.

Closed Loop Control

As written, the force control is open-loop because there is no force sensor in the PR2 arm. This creates anomalous behavior when the force being applied by the controller is not enough to overcome the static friction in the joints. At the moment, it is up to the user to work around this problem. For example, when the gripper is in contact with a surface, the gripper force sensors can be used as feedback. Although the controller itself cannot integrate this information, it can be done a higher level using the information from the "/pressure/r_gripper_motor" or "/pressure/l_gripper_motor" topics. Similarly, ensuring that the end effector moves a certain distance while constantly exerting a force in that direction (as in pushing an object) can be done by monitoring the position of the gripper and exerting ever greater forces until it begins to move. In the video below, the PR2 is using the ee_cart_imped controller to exert a light force on the table and the gripper fingertip sensors to decide whether it is currently on the table.

Video credit: Dylan Hadfield-Menell, Thomas Garver

Wiki: ee_cart_imped_control (last edited 2012-03-01 01:41:29 by JennyBarry)