Meeting Agenda for We, Oct. 27th 2pm CET

[rhaschke]

Required changes for panda_moveit_config (#183):

• Extend xacro parameterization to allow choosing:
  • hardware interface of the joint transmissions (position, velocity, or effort)
  • robot namespace in Gazebo, which also will be the namespace of published joint_states and of controller_manager
  • robotSimType for Gazebo between
    • gazebo_ros standard (gazebo_ros_control/DefaultRobotHWSim) and
    • franka_ros extension (franka_gazebo/FrankaHWSim)

Coarse collision model

• Rationale: Mimic collision model of internal controller used for self-collision

• Problem: collision model of hand is huge and doesn't allow to approach objects for grasping
  To allow grasping, one needs to disable collision of the whole hand, which is counterproductive

• Solution proposals:

  1. Use a smaller collision model for hand base + separate cylinders for fingers
     A: Collision models in Franka controller hard rather hard-coded and not easy to change.
     If we adapt the collision models, we risk having users observe self-collisions for successfully planned trajectories
  2. Alternatively, use the coarse collision model for self-collision checks only
     and add smaller models for checks with the environment
     I need to check whether that's possible in MoveIt.
     A: Weird semantics of default collision entries in ACM moveit/moveit#2936
  3. (optional) Can one disable the internal controller's checks and use the .stl meshes as proposed in #154?
     A: No.

• In any case, the collision models of the fingers should be attached to the fingers, not to the hand -> TODO

• Is it possible to adapt the safety_distance used in the internal controller, such that #153 is justified?
  A: No. Hence, it only makes sense to increase the safety_distance for the user.
  We agreed to add the 3cm default offset internally only and expose a safety distance starting from zero to the user.

Questions

• What's the difference between joint_states and joint_states_desired?
  A: joint_states_desired returns the commanded joints from previous time step. They are smoother than joint_states, which are the current values.
• Why do you use re-publishing instead of faster re-mapping?

panda_moveit_config

• Release coordination between franka_ros and ros-planning (MoveIt) was non-existent in the last year.
  Both MoveIt maintainers and Franka staff are willing to migrate panda_moveit_config back into franka_ros**
• @rickstaa and @rhaschke have put a lot of effort to get this running again after franka_ros 0.8 release
• The following operating modes are available:

  • real robot (position-controlled): roslaunch panda_moveit.launch

  • gazebo_ros sim (effort-controlled): roslaunch demo_gazebo.launch

    The effort-based hardware interface is configured by default, but other interfaces (I tested position) work as well.
    However, this requires changes in two locations: gazebo_ros_controllers.yaml and demo_gazebo.launch.

  • franka_gazebo sim (effort-controlled or Franka controllers): roslaunch panda_moveit_config franka_gazebo.launch

    To use position- or velocity-based hardware interfaces, Adds position and velocity motion generators to FrankaHWSim frankaemika/franka_ros#181 is needed

  • moveit_sim_controller (fake ROS control, no Gazebo):

    roslaunch panda_moveit_config demo.launch moveit_controller_manager:=ros_control
    roslaunch panda_moveit_config moveit_sim_controller.launch