noetic-devel fixups

[rhaschke]

See individual commits. Requires frankaemika/franka_ros#188.

[rhaschke]

franka_gazebo tests

• finger joints move in sync when setting target pose in MoveIt
• finger joints move in sync when executing GripperCommands triggered from MoveIt, but action doesn't finish:

  [ERROR] ros.moveit_ros_planning.trajectory_execution_manager: Controller is taking too long to execute trajectory (the expected upper bound for the trajectory execution was 4.886437 seconds). Stopping trajectory.
  [ INFO] ros.moveit_simple_controller_manager.ActionBasedController: Cancelling execution for franka_gripper
  [ INFO] ros.moveit_ros_planning.trajectory_execution_manager: Completed trajectory execution with status TIMED_OUT ...
  [ INFO] ros.franka_gazebo.FrankaGripperSim: Gripper Command Action cancelled
  [ INFO] ros.moveit_ros_planning_interface.move_group_interface: ABORTED: Timeout reached
  

• arm moves when triggered from MoveIt, but aborts (at the end) with GOAL_TOLERANCE_VIOLATED:

  [ WARN] ros.moveit_simple_controller_manager.SimpleControllerManager: Controller 'joint_trajectory_controller' failed with error GOAL_TOLERANCE_VIOLATED: 
  [ WARN] ros.moveit_ros_planning.trajectory_execution_manager: Controller handle joint_trajectory_controller reports status ABORTED
  [ INFO] ros.moveit_ros_planning.trajectory_execution_manager: Completed trajectory execution with status ABORTED ...
  [ INFO] ros.moveit_ros_planning_interface.move_group_interface: ABORTED: Solution found but controller failed during execution
  

• spawning two models independently:

  roslaunch gazebo_ros empty_world.launch world_name:=worlds/empty.world use_sim_time:=true
  roslaunch franka_gazebo panda.launch gazebo:=false arm_id:=left ns:=left y:=0
  roslaunch franka_gazebo panda.launch gazebo:=false arm_id:=right ns:=right y:=1
  

• spawning in a namespace:

  roslaunch franka_gazebo panda.launch ns:=foo controller:=joint_trajectory_controller
  ROS_NAMESPACE=foo roslaunch panda_moveit_config demo.launch load_robot_description:=false moveit_controller_manager:=simple
  

[rickstaa]

LGTM. Thanks for solving these issues! I tested everything except the panda_moveit.launch file. I will do this on Monday. Based on the code I however think there should be no problems other than the PID gains not being correct for the real robot. We however fix this when @frankaemika provides us with PID gains for the simulator and real robot.

arm moves when triggered from MoveIt, but aborts (at the end) with GOAL_TOLERANCE_VIOLATED:

I also noticed this during my tests. I tried to fix this with @gollth in frankaemika/franka_ros#173 and frankaemika/franka_ros#180. The issue however turned out to be a difficult one.

spawning two models independently:

Wow! Thanks for making this possible This is very helpful in my own research.

[rickstaa]

Based on the code I however think there should be no problems other than the PID gains not being correct for the real robot. We however fix this when @frankaemika provides us with PID gains for the simulator and real robot.

I just saw you are already working don't this https://github.com/frankaemika/franka_ros/pull/186/files.

[rhaschke]

Based on the code I however think there should be no problems other than the PID gains not being correct for the real robot. We however fix this when @frankaemika provides us with PID gains for the simulator and real robot.

The real robot will (still) work with position controllers, which doesn't require PID gains.

[rickstaa]

Based on the code I, however, think there should be no problems other than the PID gains not being correct for the real robot. ### We, however, fix this when @frankaemika provides us with PID gains for the simulator and real robot.

The real robot will (still) work with position controllers, which doesn't require PID gains.

I can test with the joint_trajectory_controller controller on Monday. My point, however, was about the fact that currently, the panda_arm_controller is started, which is tuned for the simulation:

panda_moveit_config/launch/ros_controllers.launch

Line 8 in e9dd0c0

<node name="trajectory_controller_spawner" pkg="controller_manager" type="spawner" respawn="false" output="screen" args="panda_arm_controller" />

In the future, we might want to start the effort PID controller provided by @frankaemika for the real robot. But let's wait till @frankaemika gives us with such a controller.

[rhaschke]

Actually, ros_controllers.launch and *.yaml are not used anymore in the latest proposal, because I switched to moveit_controller_manager:=simple. Thus, both in sim and real the joint_trajectory_controller should be used:

panda_moveit_config/config/simple_moveit_controllers.yaml

Line 2 in 890134d

- name: joint_trajectory_controller

started via franka_gazebo:

panda_moveit_config/launch/franka_gazebo.launch

Line 17 in 890134d

<arg name="controller" default="joint_trajectory_controller" />

For the real robot you might need to add spawners for the controllers.

[rickstaa]

I like this solution! Now we provide all the required controllers but don't start them by default. All this control logic is now in the franka_control package. Users can then choose the controller they want to use. I will test on the real robot.

[rhaschke]

Now we provide all the required controllers but don't start them by default. All this control logic is now in the franka_control package. Users can then choose the controller they want to use.

Yes, all controller configs are loaded as ROS params. But MoveIt's SimpleCtrlMngr requires the joint_trajectory_controller to be loaded and started. Otherwise:
[ERROR] ros.joint_trajectory_controller.joint_trajectory_controller: Can't accept new action goals. Controller is not running.

Only, the ROSCtrlMngr can start/stop controllers as well.

[rickstaa]

@rhaschke Good point! I didn't think of that. I will also test that and feedback the results to you. Feel free to join me on TeamViewer when I have all the tests set up.

[rickstaa]

@rhaschke I tested the franka_control.launch script on the real robot and it works (see gif). Based on my experiments and #14 (comment) I would like to suggest automatically starting the joint_trajectory_controller (see #15).

[rickstaa]

Maybe we can start the joint_trajectory_controller (i.e. effort-based) by default while the joint_position_trajectory_controller is only loaded by the franka_control package. By doing this, we would adhere to @frankaemika's desire to use effort-based controllers as the default. This would require only a small change in https://github.com/frankaemika/franka_ros/pull/188/files.